AI Search: How to Build Visibility in AI-Powered Search

What is AI search?

AI search is the use of generative AI within search and discovery experiences to interpret a person's request, retrieve relevant information and assemble an answer or exploration path rather than simply presenting a ranked list of links.

Google AI Overviews and AI Mode are examples within Google Search. ChatGPT and Claude can search the web and return answers with cited sources, while Perplexity, Microsoft Copilot and other systems offer their own versions of AI-assisted discovery, each with different ways of finding, combining, and presenting information. [1][2][3]

It's useful to distinguish AI search from a general response produced by a language model.

A chatbot can sometimes answer using information learned during training without searching the web again. AI search works differently: it can retrieve current external information, use that information to support its responses, and cite relevant sources. [2]

This distinction matters because much of the work discussed in this guide—crawlability, retrieval, citations, current information, and search visibility—relates specifically to experiences that use current or externally retrieved information.

Traditional search vs. AI search

AI search doesn't replace everything that came before it. Traditional search already includes rich results, videos, products, maps, direct answers and other features beyond blue links. AI-search systems can also draw on many of the same web pages and search infrastructure.

But the experience can differ in several important ways.

How AI search works

AI search systems don't all work the same way, but you can think about the path from a person's question to an answer through five broad stages. Each stage helps explain where your information can enter the process, how it might be used, and what can happen next.

Question → Retrieval → Evidence → Answer → Action

1. Question: Someone starts with a need

   They might ask:

   How can I prevent stockouts across several retail locations?

   Or:

   What inventory software is best for a retailer with five stores and a Shopify site?

   AI interfaces allow people to include more context in a single request than they often would in a short search query.

2. Retrieval: The system looks for information that may help answer the question.

   That can involve more than one search.

   Google describes query fan-out, for example, as generating several related queries to retrieve information across relevant subtopics. [4]

   A retailer asking which inventory software is best for several stores and a Shopify site might raise related questions about multi-location inventory, e-commerce integrations, stock synchronization and available software options.

   AI Mode and AI Overviews can both use query fan-out, though Google says each may use different models and techniques, producing different responses and supporting links. [1]

   The important lesson isn't that every AI system uses Google's exact method. It's that:

   One customer question can trigger a much broader retrieval process than a single keyword on a search results page.

3. Evidence: The system retrieves information that may help construct the response.

   Depending on the experience, that could include company websites, educational articles, product information, documentation, reviews, publishers, community discussions, videos, business information, structured product data and other web resources.

   Different sources may contribute different pieces of information to the system's response.

4. Answer: The system uses the information it has gathered to assemble a response.

   Depending on the question, that response might explain a concept, summarize several sources, compare alternatives, cite supporting pages, mention companies, recommend products, surface images or videos, or suggest a next step.

   Importantly, the sources used to support an answer are not necessarily the same companies or products that appear in it. A website can contribute evidence without its brand becoming part of the answer itself.

   In one 2026 Semrush study covering 3,981 domain appearances across 115 prompts, 14 countries and four AI-search experiences, 61.7% of cited appearances did not also include a brand mention. The study is observational and platform-specific, but it illustrates why citations and brand mentions should not be treated as interchangeable measurements. [5]

5. Action: What happens afterward can vary even more.

   Someone might click a cited source, ask a follow-up question, search for one of the brands mentioned, compare the recommendation elsewhere, visit the company later, make a purchase, contact sales or do nothing at all.

That brings us to the three outcomes an AI-search program needs to distinguish.

Source visibility, brand visibility, and business impact

AI visibility becomes easier to understand when you separate three different questions.

1. Source visibility: Is your information being used?

   Your page might appear as a citation, provide supporting evidence, surface for relevant retrieval queries or appear alongside an AI-generated answer.

   That tells you something about how visible your information is. It doesn't necessarily tell you whether anyone's noticed your company.

2. Brand visibility: Is your company actually part of the answer?

   Your brand might be named, described, compared, included in a shortlist, recommended, or associated with a particular problem or category.

   A company can appear as a source without being mentioned as a brand. The reverse can happen too: an AI system may mention the company in its answer without citing the company’s own website.

3. Business impact: Does any of this contribute to an outcome that matters?

   Depending on the business, that might mean increased awareness, branded search, website visits, product discovery, qualified inquiries, registrations, purchases or revenue.

   Keep these layers separate when you measure performance.

   A citation is not a recommendation. A recommendation isn't a visit. A visit is not a conversion.

AI search isn't one system

Google AI Overviews, Google AI Mode, ChatGPT, Microsoft Copilot, Perplexity, Gemini, and other experiences use different retrieval methods, source sets, models, context, answer formats, citation systems, freshness signals, geographic inputs, and measurement approaches.

Even two experiences from the same company can behave differently. Google explicitly says AI Mode and AI Overviews may use different models and techniques, resulting in different responses and supporting links. [1]

A small 2026 Semrush experiment also found substantial differences in the domains cited when the same 100 prompts were run through different ChatGPT reasoning modes. Only 25.6% of cited domains overlapped between the two test conditions. The study is too small to establish a universal rule, but it illustrates why a single prompt-response pair shouldn't be treated like a stable search ranking. [6]

There is no single universal AI-search ranking position.

Visibility needs to be evaluated across the systems, questions, and audiences that actually matter.

AI search isn't limited to text answers

Google says its generative Search features can surface images and video alongside web information and recommends supporting useful text with relevant visual content where appropriate. It also uses Merchant Center and Business Profile data in relevant product and local experiences. [4]

In September 2026, Google added dedicated multimodal reporting to Search Console for searches initiated through experiences such as Lens, Circle to Search, image uploads, and image-based Chrome searches. [7]

Commerce can go further. ChatGPT can surface products using first- and third-party product information, and some eligible products can be purchased through Instant Checkout without leaving the platform. [8]

This doesn't mean every business needs a separate strategy for every search experience. It means the information you make available should reflect how customers can actually discover your business.

For a retailer, that may include accurate product feeds. For a local business, it may mean current location and business information. For a visual category, images or video may matter. For a B2B software company like Batchify, useful web content, documentation, product information, and credible market evidence may matter most.

SEO, GEO, AEO, LLMO, and AI SEO: What's the difference?

AI search has produced a growing set of overlapping terms.

These terms can be useful for describing different aspects of the problem, but they aren't standardized and shouldn't obscure the underlying job.

Google treats optimization for its generative Search features as part of SEO and says its existing search ranking and quality systems remain foundational. [4]

While not every AI platform behaves like Google, these labels shouldn't be treated as separate marketing programs.

The practical question remains:

Where are the people we care about looking for information, and what would make our information and brand useful there?

How to improve AI search visibility in 7 steps

A practical AI-search program follows seven steps, from defining where visibility matters to measuring what changes.

Together, they form a cycle:

Objective → Questions → Baseline → Access → Evidence → Environment → Results

AI search in practice: Batchify

Bruno is CMO at Batchify, an inventory and order-management platform for multi-location retailers and ecommerce businesses.

Batchify already has a mature organic-search program. Its website includes product and feature pages, integration pages, inventory-management guides, operational templates, comparison content, and help documentation. Organic search already brings relevant visitors to the site.

SEO is still working. The change is happening earlier in the customer journey.

Prospects increasingly arrive after asking AI assistants about their inventory problems, possible approaches, and software options. When Bruno looks more closely at those answers, he finds a pattern:

Batchify sometimes appears as a source for inventory questions, but the brand appears much less often when buyers ask which software they should consider.

We'll follow Bruno's investigation throughout the seven steps.

Step 1: Define where AI search matters

Before measuring hundreds of prompts or spending on an AI-visibility platform, decide what you actually want to influence.

"Improve our ChatGPT visibility" isn't a business objective. Neither is "Rank in AI Mode."

Start with the same questions you'd ask in any serious search or content program:

• Who are we trying to reach?
• What are they trying to accomplish?
• Which parts of their journey matter to us?
• Where might AI-assisted discovery influence that journey?
• What outcome would matter to the business?

Start with the audience

A topic tells you what someone is searching about, not what they need from it. For example:

• Someone looking for a quick definition may ask:

  What is safety stock?

• A retail operator facing an immediate problem may ask:

  Why do our online stock levels keep disagreeing with our stores?

• A prospective buyer may ask:

  What inventory system should I use for five stores and Shopify?

All three questions concern inventory management, but each reflects a very different need.

The first can still build awareness and demonstrate expertise. The latter two are much closer to the problems Batchify wants to help solve and the decisions it wants to influence.

The point is to know which needs matter most to the objective you're trying to achieve.

Identify where AI could influence the journey

Consider where people in your audience might use AI to:

• understand a problem
• investigate possible causes
• compare approaches
• research categories
• evaluate products
• shortlist vendors
• answer implementation questions
• complete a task

The goal is to identify the moments where AI-assisted discovery could meaningfully shape what your audience understands, considers, or chooses.

Connect visibility to a business objective

AI visibility only matters if it supports something the business actually wants to accomplish.

An objective, like:

Get mentioned more often.

doesn't explain who you want to reach, where visibility matters, or what you hope that visibility contributes to.

A useful AI-search objective should connect the audience, the relevant part of their journey, and the business outcome you care about.

Example: Bruno defines Batchify's objective

Bruno identifies the audience Batchify most wants to reach as:

Retailers operating three or more stores or warehouses alongside e-commerce channels.

These businesses face problems such as:

• mismatched inventory
• overselling
• stockouts
• slow replenishment
• cross-location transfers
• fragmented systems

He also identifies the parts of the journey where AI search may matter most:

1. understanding the operational problem
2. deciding whether dedicated inventory software is necessary
3. evaluating possible approaches
4. comparing vendors
5. choosing a platform

Batchify's objective becomes:

Increase Batchify's visibility when multi-location retailers research inventory-management problems and evaluate relevant software, while measuring whether that visibility contributes to qualified discovery.

Now visibility has a purpose.

Step 1 output: A defined audience, a relevant part of the discovery journey, and a business objective for AI-search visibility.

Step 2: Map the questions people actually ask

Once you know where AI search matters, map the questions that represent that journey. The goal is to understand the question territory your audience moves through.

Organize questions around needs

Start with what the person is trying to accomplish.

This keeps the focus on what the person is actually trying to accomplish, rather than the exact words they happen to use.

Why not just track thousands of prompts?

Prompt variation is practically infinite.

A customer could ask:

Best inventory software for five stores.

Or:

I run four stores plus a Shopify site and keep overselling online because our locations aren't synchronized. What software should I look at?

The wording is very different, but the underlying decision is much the same.

Google's query fan-out model illustrates why this matters. One complex request can produce multiple related retrieval queries behind the scenes. Creating separate pages for every possible prompt variation is neither necessary nor a sound content strategy. Google explicitly advises against producing separate pages primarily to capture every query or fan-out variation. [4]

Instead, think in terms of needs, topics, and decision territory, and treat individual prompts as samples within that territory.

Choose a representative question set

You still need specific questions to measure. Build a set that reflects:

• important audience needs
• different stages of the journey
• meaningful product and category questions
• major customer situations
• language real customers might use

Don't choose questions simply because your brand already appears for them, and don't treat the set as a perfect representation of demand.

The goal is to create a consistent measurement set you can use to understand what happens across the territory that matters.

Example: Bruno maps Batchify's question territory

Bruno creates 24 representative questions: four for each of the six needs. Here are two examples from each group:

Learn: Someone is trying to understand a concept

• What is safety stock?
• What is inventory turnover?

Solve: Someone is trying to fix a problem

• How do I stop overselling across my online store and retail locations?
• How can I reduce stockouts across several stores?

Compare: Someone is trying to understand trade-offs 

• Shopify inventory management vs. dedicated inventory software
• Centralized vs. location-level inventory management

Evaluate: Someone is trying to decide what matters

• What features matter in inventory software for multiple stores?
• When does a retailer need dedicated inventory-management software?

Choose: Someone is trying to select an option

• Best inventory software for a five-store retailer
• Best inventory platform for Shopify plus physical stores

Use: Someone is trying to complete or improve a task

• How should inventory software integrate with a POS?
• How should retailers manage transfers between locations?

The full 24-question set isn't meant to represent the entire market. It gives Batchify a consistent sample across the customer journey.

Visibility can vary even across closely related questions. A 2026 Semrush study of more than 1,000 subject areas in ChatGPT found considerable variation in brand visibility across related prompts, reinforcing the value of measuring a topic through multiple questions rather than treating one response as representative. [9]

Step 2 output: A question map organized around meaningful audience needs and a consistent question set for measurement.

Step 3: Establish your AI-search visibility baseline

Now establish what happens when relevant AI systems answer those questions.

Simply asking “Are we there?” isn't enough. You also need to understand how you appear.

Separate citations from brand mentions

For each representative question, record whether:

• your site is cited
• your brand is mentioned
• your product is compared or recommended
• relevant competitors are mentioned
• important third-party sources appear
• the answer describes your company accurately
• the result differs materially across platforms

Each observation tells you something different about your visibility.

Microsoft's AI Performance report makes a similar distinction at the platform level. Bing reports citation activity and grounding queries, but explicitly says citations do not represent rankings, authority, traffic, or engagement. [10]

Measure across the platforms that matter

Choose the AI-search experiences that are most relevant to your audience.

Depending on the business, that might include:

• Google generative search
• ChatGPT Search
• Claude with web search
• Microsoft Copilot
• Perplexity

Keep the results separate where possible. A combined "AI visibility score" can be useful as a summary, but it may also hide meaningful differences in where and how your company appears across platforms.

Record enough context to make comparisons meaningful

AI-search results can vary across platforms, markets, modes, and measurement periods.

When practical, record:

• question
• platform
• date
• market or location
• relevant mode or product experience
• whether web search was used
• source citations
• brand mentions
• competitor mentions
• notable answer characteristics

The goal is to create a consistent enough record so that changes mean something when you return to the same questions later.

Use first-party data where available

Platform reporting can add another layer to your baseline.

Google now provides dedicated generative-AI performance reporting in Search Console, including generative-feature impressions, pages, countries, devices, and dates. The reports were rolled out globally on August 31, 2026. [11]

Microsoft's Bing Webmaster Tools also provide citation activity and grounding-query information across supported AI experiences. [10]

These measurements are especially useful because they come directly from the platforms.

Third-party AI-visibility tools can provide broader cross-platform monitoring, but their results depend on their own prompt sets, locations, models, sampling methods, and methodologies.

Example: Bruno establishes Batchify's baseline

Bruno runs Batchify's 24 representative questions across three relevant AI-search experiences. That produces 72 platform-question observations.

The initial pattern looks like this:

All figures are illustrative. Citation, mention, and competitor columns can overlap within the same response.

The pattern tells Bruno far more than a single visibility score would.

Batchify appears reasonably often as a source for educational and operational questions, but its brand appears much less frequently. The gap is largest when buyers move into comparison, evaluation, and selection.

Bruno's diagnosis:

Batchify's information contributes to some AI answers, but that source visibility rarely translates into brand visibility when prospective buyers evaluate software.

Now he has a specific problem to investigate.

Step 3 output: A repeatable baseline showing where your information, brand, and competitors appear across the questions and platforms that matter.

Step 4: Make your information discoverable

AI-search optimization starts with technical eligibility. Before changing content, make sure the systems that matter can access it.

Start with search fundamentals

Check whether important pages are:

• publicly accessible
• crawlable
• indexable
• correctly canonicalized
• linked internally
• available in usable textual form
• technically stable
• current where freshness matters

For Google, there is no separate technical path into AI Overviews or AI Mode. A page first needs to be indexed and eligible to appear in regular Google Search. If it isn't eligible to appear there, it won't appear in Google's generative Search features either. [1]

Understand platform-specific crawler controls

Different platforms provide different controls over how their systems access your site.

Both OpenAI and Anthropic distinguish between crawlers used for search discovery, model training, and other forms of web access.

For OpenAI:

• OAI-SearchBot is used to surface websites in ChatGPT Search.
• GPTBot crawls content that may be used to improve OpenAI's generative AI foundation models.

For Anthropic:

• Claude-SearchBot crawls the web to support Claude's search results.
• ClaudeBot collects web content that may contribute to model training.

Both companies allow site owners to control these purposes separately. That means you can restrict access for model training without necessarily blocking search discovery.

However, allowing a search crawler only makes your information eligible to be retrieved. It does not guarantee that your content will appear in an AI-generated answer.

Search crawling isn't the only way AI assistants can access a page

AI assistants can also retrieve web content in response to a user's request.

OpenAI documents ChatGPT-User, which may visit a page as part of certain user-initiated actions in ChatGPT. Anthropic similarly documents Claude-User, which may access a website when someone asks Claude a question that requires its content. These serve a different purpose from the search crawlers each company uses to discover or index information in advance. [12] [13]

This becomes increasingly important as AI assistants and agents browse the web to research questions, compare options, check current information, or complete tasks on someone's behalf. In those cases, software may be the immediate reader even though a human is still the ultimate beneficiary.

Important information should therefore be available in forms automated systems can reliably retrieve and interpret.

Create information for people, then make sure machines can access and understand it.

A page that's technically easy for an AI system to read but unhelpful, generic, or unsupported is still weak information.

Do you need llms.txt?

llms.txt is a proposed text file intended to make information easier for language-model systems to consume.

It may become useful within particular systems or workflows, but it should not currently be treated as a universal AI-search requirement. Google explicitly says it doesn't use llms.txt for Google Search and that creating one neither helps nor harms Search visibility. [4]

A 2026 Ahrefs study also found that 97% of the roughly 38,000 valid llms.txt files it observed received no requests during its measurement period. That does not prove the format will never become useful, but it provides little support for treating it as an important current visibility lever. [14]

If another system you use explicitly supports the file, that may be a reason to maintain one.

What about schema?

Schema is structured data added to a webpage's code to help search engines understand what the page contains. It can identify things like products, organizations, reviews, events, and other types of information, and support eligible search features.

But schema isn't a requirement for visibility in Google's AI search features. There is no special schema type for generative Google Search, and Google explicitly says structured data is not required for inclusion in its AI features. [4]

A 2026 matched-page study by Ahrefs also found no meaningful increase in AI citations after pages added JSON-LD schema across the systems it tested. That does not prove schema can never matter in any context. It simply does not support treating schema implementation as a demonstrated AI-citation tactic. [15]

Make the right information available for your business model

What needs to be discoverable depends on the business.

An e-commerce site may need accurate information about:

• products
• availability
• pricing
• feeds
• product metadata

A local business may depend on current information such as:

• address
• opening hours
• service information
• Business Profile data

A publisher or software company may rely more heavily on:

• accessible pages
• documentation
• articles
• comparison information
• product details

Visual content can matter too. Google specifically recommends relevant images and video for generative Search where they improve the user experience. [4]

The goal is to make the information your audience needs available, accurate, and accessible wherever discovery happens.

Example: Bruno checks Batchify's technical eligibility

Bruno's team reviews the pages associated with Batchify's priority questions.

They find that:

• important pages are indexed
• relevant content is crawlable
• no major robots rules prevent access
• OAI-SearchBot is permitted
• canonicals are correct
• product and documentation pages are internally linked
• important information is available as normal page content

The team does not have an llms.txt file, but Bruno decides that's not a priority. Access doesn't appear to be Batchify's main bottleneck. Its information is already being retrieved for some of the questions that matter.

The larger issue is what information exists and how well it helps buyers make more commercial decisions.

Step 4 output: A clear view of whether technical access is limiting AI-search visibility, plus any issues that actually need fixing.

Step 5: Create information worth using as evidence

Once you've made your information discoverable, the next question is whether your content is useful enough to support an answer.

Start with the person's question

The question territory we established earlier reflects what someone is trying to accomplish: learn, solve, compare, evaluate, choose, or use.

Bring that same lens to the information itself. Someone trying to learn may need a clear explanation and examples. Someone trying to solve a problem may need practical instructions or evidence. Comparison and evaluation questions may require criteria, trade-offs, data, and expert judgment. Someone ready to choose may need current product information, evidence of fit, and clear limitations. A person trying to use something may need detailed instructions, examples, or documentation.

For each important question territory, ask:

What would genuinely help someone make progress here?

A useful answer may combine several of these elements. What matters is whether the information helps someone understand, solve a problem, or make a decision.

Commodity information is particularly weak

Generative systems are already good at synthesizing common information. That makes undifferentiated content especially easy to replace.

Google's current guidance explicitly recommends non-commodity content: material with original experience, unique perspective, expertise, and information beyond what can simply be restated from elsewhere. [4]

A useful principle is:

If an AI system can produce the same information without you, reproducing that information more neatly is not an advantage.

That doesn't mean every page needs proprietary research.

A clear, useful explanation can still be valuable. The question is what your resource contributes that isn't interchangeable with dozens of others.

Make claims supportable

If you make factual or comparative claims, give people—and machines—a way to understand where they came from.

Useful support may include:

• original data
• cited external research
• methodology
• clear examples
• first-hand experience
• transparent product information

Avoid invented specificity or unsupported claims just to make a claim sound authoritative.

Help people understand important distinctions

Despite a common assumption, content does not need to be broken into small "AI chunks" to perform in AI search. Instead, organize complex information so the differences are easy to understand.

For example, a useful Batchify resource might clearly distinguish between:

• inventory software and warehouse-management software
• native e-commerce inventory and dedicated inventory systems
• real-time sync and scheduled reconciliation
• inventory planning and inventory tracking

Google explicitly says there is no requirement to break content into small chunks for its generative features and recommends writing for the audience rather than optimizing page length or structure for AI systems. [4]

Improve existing resources before creating new ones

AI-search enthusiasm can quickly turn into unnecessary publishing. Before creating a new page, ask whether an existing resource already serves the need and whether it could do that job better:

• Does an existing page already address this need?
• Is the information accurate and current?
• Does it answer the important questions?
• Could the explanation, evidence, or structure be improved?
• Would a new page substantially overlap with something you already have?
• Is there a genuinely distinct information need that deserves its own resource?

For an existing resource, the right intervention may be to keep, update, expand, simplify, reposition, consolidate, or remove it. If the need is genuinely distinct and no existing resource serves it well, creating something new may be justified.

This is a content optimization decision: diagnose what is preventing the existing resource from doing its job before deciding that more content is the answer.

Example: Bruno compares Batchify's strong and weak resources

Batchify already has a detailed guide to calculating safety stock. It explains the variables, includes worked examples, discusses where simple formulas break down, and provides a spreadsheet template.

That page appears frequently as a cited source in Bruno's baseline.

Batchify's main inventory-software page is different. It describes the product's features, but does little to help a retailer answer the questions that come before choosing a system.

Bruno identifies several gaps:

Batchify doesn't need another Best Inventory Management Software in 2026 article. It needs a stronger existing resource that helps the intended buyer make an informed decision.

Bruno's team expands and improves the inventory-software page with:

• a decision framework for when dedicated software becomes worthwhile
• a clear explanation of multi-location complexity
• the integration requirements buyers should consider
• trade-offs among different approaches
• realistic implementation considerations
• scenarios where a lighter solution may be sufficient
• transparent information about which retailers Batchify serves best

The improved resource is more useful to buyers and gives AI systems better information to work with when those questions arise.

Step 5 output: A prioritized set of resources to keep, update, expand, simplify, reposition, consolidate, or remove, with new resources created where a genuinely distinct need exists.

Step 6: Understand the information environment around your brand

Your website is only one part of what an AI system may encounter. When people ask broader market questions, AI-search experiences can retrieve information from a wide range of sources across the web.

That can include:

• editorial publications and specialist websites
• communities and Q&A platforms such as Reddit and Quora
• social and professional platforms including Facebook, Instagram, TikTok, and LinkedIn
• video platforms like YouTube
• reference sources such as Wikipedia and dictionaries
• research, academic, government, and institutional sources
• customer-review and comparison platforms
• retailers and product marketplaces
• directories and app stores
• company and competitor websites

The exact mixture varies by platform and question. Understanding which sources shape the answers around your category and brand gives you a clearer view of the information landscape you're competing in.

Current citation data shows how varied that landscape can be. In Ahrefs' September 2026 analysis of ChatGPT citations across a broad set of US queries, Reddit was the most-cited domain, followed by Wikipedia. Other frequently cited sources included Consumer Reports, Forbes, YouTube, government and medical sites, retailers, review platforms, specialist publications, LinkedIn, and app stores. [16]

See which sources appear repeatedly

Return to the questions that matter and look for patterns:

• which external sources are cited repeatedly
• where your competitors are discussed or compared
• which communities, review platforms, and reference sources recur
• whether video or other media regularly appears
• which market claims and descriptions recur
• whether important information about your company is missing, outdated, or inaccurate

These patterns reveal where the information and evidence shaping your market comes from, and which sources influence how the category is understood.

Treat repeated appearances as signals worth investigating. Frequency alone does not establish causality, and being mentioned on a commonly cited source does not mean an AI system will cite, recommend, or surface your company.

Compare what you say with what the market says

A company can explain its own product, but claims like "best order management platform" carry more weight when supported by credible external evidence.

That evidence might come from:

• real customer experience
• independent testing
• expert analysis
• credible reviews
• documented industry use
• original research that others choose to cite

AI systems may encounter both first-party and third-party information. OpenAI's shopping documentation, for example, says its product experiences can use first-party and third-party metadata, reviews, and other external product information. [8]

Don't turn this into mention farming

Google explicitly warns against seeking inauthentic mentions as a generative-search tactic. [4]

You can't manufacture credible community mentions, and getting your brand name onto as many websites as possible is not a useful strategy.

Focus on creating accurate information, earning legitimate coverage, and giving customers, experts, and other third parties something worth discussing.

Possible actions include:

• encouraging real customers to leave honest reviews where appropriate
• making accurate product information available to relevant platforms
• publishing original research that others have reason to cite
• providing useful expert commentary
• correcting material inaccuracies
• contributing meaningfully to relevant communities
• maintaining accurate listings
• creating useful videos or demonstrations
• giving independent evaluators enough information to assess the product fairly

The strongest signals come from credible evidence that deserves to exist.

Example: Bruno investigates Batchify's information environment

Bruno notices several patterns across the sources appearing in Batchify's comparison, evaluation, and selection questions.

AI answers frequently draw from:

• software review sites
• ecommerce publications
• retailer communities
• YouTube comparisons
• independent consultants
• vendor comparison pages

Across those sources, several competitors appear more consistently than Batchify.

The team also finds clear information gaps.

Some comparison pages still describe Batchify primarily as an e-commerce inventory tool, despite its support for physical retail operations. Several review profiles contain outdated screenshots or incomplete feature information.

Batchify has strong customer stories on its own site, but relatively few independent accounts of its multi-location use cases. The company also has operational data that could support useful industry research but has never published it.

Bruno's team decides to:

• correct outdated third-party product information where possible
• make current product details easier for reviewers and partners to verify
• publish an anonymized study of inventory discrepancies across multi-location retailers
• expand YouTube demonstrations around common operational problems
• ask satisfied customers for honest reviews through its normal customer program

None of these actions guarantees an AI mention. They strengthen the information available about Batchify across the places buyers and AI systems may encounter it.

Step 6 output: A map of the external sources and evidence surrounding important questions, plus legitimate information gaps worth addressing.

Step 7: Measure what changed and decide what's next

After making changes, return to your baseline and measure the same questions again. Keep in mind that AI-search results can change for reasons beyond your work. Models and retrieval systems evolve, competitors publish new information, search behavior changes, and your own website continues to develop.

The goal is to determine what changed, what remains uncertain, and what the evidence justifies doing next.

Repeat the same measurement where possible

Use the same:

• question set
• platforms
• markets
• measurement definitions

Record meaningful differences in product mode or system behavior when you know them. The more consistent the measurement, the more useful the comparison.

Compare source visibility

Return to the same question set and compare how your content appears as a source:

• Are more relevant pages being cited?
• Are citations appearing across more question groups?
• Which resources are being cited most often?
• Where has citation visibility increased or decreased?

First-party platform data can add another layer of evidence. Google's generative-AI reporting provides impression trends and page-level visibility across its AI Search features. [11]

Bing's AI Performance report also lets publishers track citation volume, cited pages, and grouped grounding queries over time, while cautioning that changes in these metrics cannot automatically be attributed to a particular update or intervention. [10]

Compare brand visibility separately

Source visibility only tells part of the story. Next, look at how the brand itself appears:

• Is the brand appearing more often?
• In which question groups?
• Is it being described accurately?
• Is it being compared with the right alternatives?
• Are important product strengths and limitations represented fairly?

An increase in citations does not necessarily mean the brand itself has become more visible. Track source visibility and brand visibility separately.

Measure what happens beyond the answer

Visibility matters most when it contributes to something meaningful for the business. Look beyond the AI response itself and examine:

• AI referral traffic
• branded search
• landing-page engagement
• registrations
• inquiries
• purchases
• other meaningful outcomes

OpenAI automatically adds utm_source=chatgpt.com to ChatGPT Search referral URLs, which can help publishers identify attributable ChatGPT referral traffic in their analytics. [17]

Not every AI-influenced journey produces a direct referral, or even a visible brand mention. Your information may contribute to an answer without the user noticing the source. Someone who does encounter your brand may later:

• search for your brand
• navigate directly to your site
• return days later
• speak to a colleague
• contact sales through another channel

That makes attribution imperfect. Be clear about what the evidence shows and where attribution becomes uncertain:

What we observed
What may have contributed

Example: Bruno measures Batchify's follow-up

Three months later, Bruno repeats the same 24-question measurement across the same three AI-search experiences.

The illustrative results look like this:

The direction is encouraging. Batchify's brand mentions increased from 10 to 18 across the 72 observations, an 80% increase in observed mentions.

It would be tempting to credit that increase to the work Bruno's team completed. The timing fits, and the largest improvement appears in the commercial question groups the team deliberately addressed.

But the measurement does not establish causality. During those three months, Batchify changed multiple resources and improved information beyond its own website. AI models and retrieval systems may also have changed, competitors may have published new information, and the same questions can produce different results over time.

Bruno can confidently describe what changed, but he can't isolate how much of that change came from the buyer guide, another Batchify intervention, or shifts in the wider search environment.

What he can say is:

Batchify is appearing more consistently across relevant evaluation questions while maintaining healthy source visibility. The clearest improvement is in the commercial question groups the team deliberately addressed, although some high-value selection questions still rarely include the company.

He also notices that Batchify's inventory study is now appearing as a source in several educational and evaluation answers.

That sparks the next question:

How can Batchify turn stronger visibility and useful market evidence into more consistent consideration across the remaining high-value questions?

And the cycle starts again.

Step 7 output: A measured result, a clear view of what remains uncertain, and evidence for deciding what to do next.

AI-search metrics at a glance

No single metric captures AI-search performance. Each answers a different question.

Be precise about AI share of voice

"AI share of voice" can sound like a standardized metric, but the result depends heavily on how it is measured.

Before comparing scores, understand:

• which platforms are included
• which questions are included
• how often those questions are rerun
• which country or market is used
• whether citations, mentions, recommendations, or some combination are counted
• whether different placements are weighted
• which model or product experience was tested

A 30% share of voice across 100 commercial questions is not directly comparable with 30% across 1,000 educational questions.

Which AI-search tools do you need?

You don't need every new AI-visibility platform to run a useful program. Different tools answer different questions.

Native platform reporting

Start with first-party data where it is available. Examples include:

• Google Search Console
• Bing Webmaster Tools

These tools provide visibility data directly from the platforms themselves, helping you understand where and how your content appears in their search experiences.

Web analytics

Use your analytics platform to understand identifiable AI referral traffic and what happens after people reach your site.

Depending on your setup, that may include:

• landing pages
• engagement
• conversions
• revenue
• identifiable referral sources

This connects AI-search visibility with measurable activity on your own site.

AI-visibility platforms

Third-party AI-visibility platforms can help you:

• monitor representative questions at scale
• track citations and brand mentions
• compare competitors
• compare platforms
• preserve historical observations

They can make ongoing monitoring much easier, but results depend on the questions, markets, models, sampling methods, and definitions they use. Their value is observational: they show you what appears in the AI-search experiences they monitor, without access to the internal ranking or generation systems behind those results.

Traditional SEO tools

Traditional SEO tools remain useful for understanding the search and web environment that AI systems may draw from.

Depending on the tool, that can include:

• search demand
• rankings
• backlinks
• technical conditions
• existing content
• competitors
• search results
• site performance

AI search adds new dimensions to measure while many of the underlying search signals, websites, and competitive dynamics remain relevant.

Manual review

Some questions still require you to read the answers yourself. A dashboard might tell Bruno that Batchify appears in 14% of tracked responses. But only reading those responses can reveal:

• whether Batchify is described accurately
• which competitors appear
• why particular options are being discussed
• which sources support the answer
• what important information is missing
• what questions a buyer might ask next

Use quantitative monitoring to find patterns and qualitative review to understand what those patterns actually mean.

Common AI-search mistakes

Common AI-search mistakes

Treating AI search as completely separate from SEO

For AI experiences grounded in web search, familiar fundamentals still matter: crawling, indexing, relevance, quality, and useful content.

Google explicitly says existing SEO fundamentals remain relevant to its generative Search experiences. [4]

Check basic access and search eligibility before looking for more complicated explanations.

Treating one prompt or platform as the market

One answer is an observation, not a market. Results can vary across related questions, platforms, modes, and measurement periods.

Measure across the question territory and systems that matter before drawing conclusions about your visibility.

Conflating citations, brand mentions, and business outcomes

A page can be cited without the company being mentioned. A brand can be mentioned without anyone visiting its website. Neither outcome establishes a conversion.

Keep source visibility, brand visibility, and business impact separate.

Creating content for prompts instead of people

Question wording can vary almost endlessly. Creating a page for every possible prompt is unlikely to produce a coherent content program.

Build resources around genuine needs and distinct question territory. Clear structure can help both people and machines, but Google explicitly says its generative Search features do not require content to be broken into small "AI chunks." [4]

Publishing more instead of improving what already exists

More content does not automatically create more visibility.

An existing resource may need to be updated, expanded, simplified, repositioned, consolidated, or left alone. Create something new when there is a genuinely distinct need worth serving.

Treating llms.txt or schema as AI-search shortcuts

Neither should be treated as a proven route to greater AI visibility.

Google says it does not use llms.txt for Search, and there is currently no broad evidence that adding the file improves AI-search visibility. [4][14]

Structured data remains useful where it supports established search features and accurate machine-readable information, but it should not be assumed to increase AI citations.

Manufacturing third-party mentions

Artificial discussion around your brand does not create credible market evidence.

Focus on accurate product information, legitimate coverage, honest customer experience, useful research, and meaningful participation in relevant communities.

Measuring visibility that does not matter to the business

A company can become more visible across AI systems while making little progress with the audience or decisions that matter.

Tie measurement back to the audience, question territory, journey, and business objective you defined at the outset.

Turning correlation into causation

Patterns are useful for generating hypotheses, but they do not automatically explain why something happened.

If frequently cited pages share a characteristic, that does not prove the characteristic caused them to be cited. Keep four things separate:

• platform documentation
• observed patterns
• hypotheses
• demonstrated causal effects

That distinction matters especially in a fast-changing field where confident claims can spread faster than reliable evidence.

Frequently asked questions

What is AI search?

AI search uses generative AI to interpret a person's request, retrieve relevant information, and produce an answer or exploration experience. Depending on the platform, that experience may include citations, links, products, images, comparisons, or other information.

What is an AI search engine?

An AI search engine is a search or discovery system that uses AI to interpret requests and generate responses rather than relying entirely on a conventional list of search results.

The category is broad. Some products are AI-native search engines, while others add generative features to established search products or conversational assistants.

What is AI SEO?

AI SEO is an informal term that can refer to either using AI to perform SEO work or improving visibility within AI-powered search experiences.

Because the term is ambiguous, it is useful to define which meaning you intend.

What is GEO?

Generative engine optimization, or GEO, generally refers to efforts to improve visibility within generative search and answer systems.

For Google specifically, the company says work intended to improve visibility in its generative Search features remains part of SEO rather than requiring a separate optimization discipline. [4]

What is AEO?

Answer engine optimization, or AEO, generally refers to improving how useful and visible information is within systems that answer users' questions directly.

There is substantial overlap among AEO, GEO, SEO, and AI-search optimization.

What is LLMO?

LLM optimization, or LLMO, is a broad term for efforts intended to improve how information or brands appear in experiences powered by large language models.

It is not a standardized technical discipline or a universal ranking system.

Is GEO different from SEO?

The distinction depends on the platform and how the terms are being used.

Traditional SEO focuses on organic search visibility. GEO typically emphasizes generative answers, citations, mentions, and AI-assisted discovery.

In practice, much of the underlying work overlaps, particularly where AI experiences rely on conventional web-search indexes and ranking systems.

Google explicitly treats optimization for its generative Search experiences as part of SEO. [4]

How do you optimize for AI search?

Start with the audience and questions that matter.

Then:

1. establish your current visibility
2. make important information accessible
3. create useful, differentiated resources
4. understand the wider information environment around your brand
5. measure citations, mentions, traffic, and business outcomes separately
6. use the results to decide what to improve next

Avoid assuming there is one universal formula for AI-search visibility.

How do you appear in Google AI Overviews?

Google says pages must be indexed and eligible to appear in normal Google Search with a snippet. There are no additional technical requirements specifically for appearing in AI Overviews or AI Mode. Existing SEO fundamentals and useful, people-first content remain applicable. [1]

How do you get cited by ChatGPT?

There is no guaranteed method.

For ChatGPT Search specifically, OpenAI recommends allowing OAI-SearchBot so your site can be considered for search results. [12]

Beyond accessibility, focus on providing accurate, useful information that serves the questions your audience actually asks.

Eligibility does not guarantee citation.

What is query fan-out?

Query fan-out is a retrieval technique in which an AI system generates several related searches from a broader user question.

Google says AI Mode and AI Overviews can use query fan-out to gather information across related subtopics and data sources. [1][4]

Does llms.txt improve AI-search visibility?

There is currently no evidence that llms.txt is a universal AI-search requirement.

Google explicitly says it does not use llms.txt for Google Search and that the file neither helps nor harms visibility there. [4]

Other systems may support the format now or in the future, so evaluate it according to the platforms and workflows you actually use.

Does schema help AI search?

Structured data can still be valuable for established search features and for communicating supported machine-readable information.

However, Google says no special schema is required for its generative Search experiences. [4]

Use schema where it is genuinely supported, not as a presumed AI-citation tactic.

How do you track AI-search visibility?

Use a combination of:

• first-party platform reporting
• representative question monitoring
• citation tracking
• brand-mention tracking
• AI share of voice
• referral traffic
• website outcomes

Measure groups of related questions rather than treating one prompt as representative of your overall visibility.

Are AI citations the same as backlinks?

No. A traditional backlink is a link published on another webpage. An AI citation is a reference or source surfaced within an AI-generated answer.

Both may direct someone to your site, but they are created and measured differently.

Microsoft explicitly says its AI citation metrics do not represent ranking, authority, traffic, or engagement. [10]

Final thoughts: Be worth finding

AI search changes how information is discovered, assembled, and presented. The underlying job remains the same: help people understand, solve, compare, evaluate, choose, and act.

Start there.

Understand the questions that matter. Make the right information accessible. Create resources worth using. Strengthen the evidence around your brand. Measure what changes, then decide what deserves attention next.

Be useful when your audience needs an answer, easy to find when they look, and credible when your business belongs in the answer.

Sources

• [1] Google's Guide to Optimizing for Generative AI Features on Google Search | Google Search Central  |  Documentation  |  Google for Developers developers.google.com ↗
• [2] AI Features and Your Website | Google Search Central  |  Documentation  |  Google for Developers developers.google.com ↗
• [3] Introducing Search Generative AI performance reports in Search Console  |  Google Search Central Blog  |  Google for Developers developers.google.com ↗
• [4] Searching the web with ChatGPT | OpenAI Help Center help.openai.com ↗
• [5] Overview of OpenAI Crawlers developers.openai.com ↗
• [6] Publishers and Developers - FAQ | OpenAI Help Center help.openai.com ↗
• [7] AI Performance - Bing Webmaster Tools bing.com ↗
• [8] Shopping with ChatGPT Search | OpenAI Help Center help.openai.com ↗
• [9] AI Visibility Is a Topic-Level Game — Semrush semrush.com ↗
• [10] AI Performance in Bing Webmaster Tools — Microsoft Bing bing.com ↗
• [11] Introducing Search Generative AI Performance Reports in Search Console — Google Search Central developers.google.com ↗
• [12] Overview of OpenAI Crawlers — OpenAI developers.openai.com ↗
• [13] Does Anthropic crawl data from the web, and how can site owners block the crawler? — Anthropic Help Center support.claude.com ↗
• [14] We Analyzed 137K Sites: 97% of llms.txt Files Never Get Read — Ahrefs ahrefs.com ↗
• [15] We Tracked 1,885 Pages Adding Schema. AI Citations Barely Moved — Ahrefs ahrefs.com ↗
• [16] The 50 Most-Cited Websites in ChatGPT (September 2026) — Ahrefs ahrefs.com ↗
• [17] Publishers and Developers - FAQ — OpenAI help.openai.com ↗