AI Visibility: Get Cited by ChatGPT, Perplexity and AI Overviews

What is AI visibility?

AI visibility is how often, and how prominently, your site is cited, linked or named in answers from AI systems: Google AI Overviews and AI Mode, ChatGPT search, Perplexity, Microsoft Copilot and Claude. It is the generative-answer counterpart to organic rankings, measured in citations and mentions rather than positions.

This is a tutorial, not a theory piece. It works through four steps in the order they fail: be eligible, be the best source, be mentioned elsewhere, and measure. Each step is built from the platforms’ own documentation, with the evidence labeled, because this field has more confident advice than confirmed facts. For the background on where the techniques come from, read what is generative engine optimization; for the definition alone, the AI visibility glossary entry.

How each platform finds and cites sources

Every one of these systems retrieves from an index and then writes. The differences that matter to you are which index, which crawler, what the crawler does with robots.txt, and how the citation is shown. This table is assembled from each company’s documentation; links follow it.

Platform	Index it draws on	Crawler or fetcher	robots.txt behavior	How you are cited
Google AI Overviews and AI Mode	Google’s Search index; the page “must be indexed and eligible to be shown in Google Search with a snippet”	Googlebot. No separate AI crawler. Google-Extended is a token for Gemini training and grounding and “does not impact a site’s inclusion in Google Search”	Googlebot rules apply. nosnippet and max-snippet also limit use as a direct input to AI features	Supporting links shown beside or within the response; AI Mode links inline
ChatGPT search	OpenAI’s own crawl. “OAI-SearchBot is for search”	OAI-SearchBot (search index), ChatGPT-User (fetches a page when a user asks), GPTBot (training)	OAI-SearchBot honors it; opted-out sites “will not be shown in ChatGPT search answers”. ChatGPT-User: “robots.txt rules may not apply”	Inline citations with a sources list on the response
Perplexity	Perplexity’s own index, built by PerplexityBot	PerplexityBot (index), Perplexity-User (fetches when a user asks)	PerplexityBot: Perplexity recommends allowing it. Perplexity-User “generally ignores robots.txt rules”	Numbered inline citations with source cards
Microsoft Copilot and Bing AI summaries	Bing’s index; Copilot sends a generated query “via secure connection to the Bing search service”	bingbot	bingbot rules apply. data-nosnippet content is “excluded from snippets and AI summaries”	Numbered references linked to source pages
Claude	Anthropic’s search index plus live fetches	Claude-SearchBot (index), Claude-User (fetches when a user asks), ClaudeBot (training)	All three honor robots.txt “do not crawl” signals	Linked citations in the response

Sources: Google AI features, OpenAI bots, Perplexity bots, Anthropic crawlers.

Two implications shape everything below. Google and Microsoft reuse their search indexes, so ordinary indexing in Google and Bing is most of the eligibility work. OpenAI, Perplexity and Anthropic run their own crawlers with their own robots.txt names, so a robots.txt written for Googlebot can silently exclude you from three engines. Our AI crawlers guide lists every bot with its purpose.

Step 1: Be eligible

Eligibility is binary and it is where most sites fail without noticing. Nothing in steps 2 to 4 matters until a page passes all four checks here.

1.1 Allow the search bots in robots.txt, and decide separately about training

Open your robots.txt and check each of the five search bots by name. A rule for * applies to them unless a more specific group exists, and a blanket block added to keep training crawlers out often catches the search crawlers too.

# Search and answer crawlers: allow
User-agent: Googlebot
User-agent: Bingbot
User-agent: OAI-SearchBot
User-agent: PerplexityBot
User-agent: Claude-SearchBot
Allow: /

# Training crawlers: your policy choice.
# These settings do not affect search inclusion on any platform.
User-agent: GPTBot
User-agent: ClaudeBot
User-agent: Google-Extended
Disallow: /

OpenAI’s documentation says each bot’s setting “is independent of the others” and gives the example of allowing OAI-SearchBot while disallowing GPTBot; changes take about 24 hours to take effect in ChatGPT search. Anthropic’s bots “respect ‘do not crawl’ signals by honoring industry standard directives in robots.txt”. Google-Extended controls Gemini training and grounding only. Whether you block the training crawlers is a policy decision; it is not a visibility one. Test any URL against every bot with the robots.txt tester, and see the glossary entry for OAI-SearchBot if the name is new to you.

1.2 Be indexed in Google and in Bing

Google says the only technical requirement for AI Overviews and AI Mode is that the page is indexed and snippet-eligible: “There are no additional requirements to appear in AI Overviews or AI Mode, nor other special optimizations necessary.” Confirm with URL Inspection in Search Console.

Bing is the index most sites ignore and the one Copilot reads. Verify the site in Bing Webmaster Tools, submit the sitemap, and check the page’s index status there. Microsoft recommends IndexNow so that updates reach Bing immediately and “AI systems reference the most current version of your content”. Validate the sitemap with the sitemap checker first; a sitemap with broken entries is a slow way to be indexed anywhere.

1.3 Check snippet eligibility

A page that cannot show a snippet cannot be cited. Search the HTML and headers for noindex, nosnippet, a low max-snippet, and data-nosnippet wrappers around the content you want quoted. Google’s robots meta documentation says nosnippet and max-snippet “also limit how much of the content may be used as a direct input for AI Overviews and AI Mode”; Bing applies data-nosnippet to “snippets and AI summaries”. These controls are useful when you want them. Audit them so you only have the ones you chose.

1.4 Make the content extractable without a click

Disable JavaScript and reload the page. If the answer disappears, it may not exist for some retrieval systems. Then look for tabs, accordions and “read more” controls; Microsoft’s guidance for AI answers says systems “may not render hidden content” and tells publishers not to hide important answers in tabs or expandable menus. Google asks that important content “is available in textual form”. Render the text in the initial HTML and let CSS handle the folding if you must fold.

The free SEO audit runs checks 1.1 to 1.4 for a URL in about a minute, including a fetch of the page as each AI bot to compare what they receive.

Step 2: Be the best source for the question

Once a page can be retrieved, the engine chooses which retrieved pages to quote. Evidence on what wins here comes from one controlled benchmark and from the engines’ own advice, and the two agree.

2.1 Target the questions, including the ones the engine will generate

Google says AI Overviews and AI Mode “may use a ‘query fan-out’ technique, issuing multiple related searches across subtopics and data sources”. The practical move is to answer the sub-questions on the page, each under its own heading, so the page is retrievable for a query the user never typed. Find the sub-questions from People Also Ask boxes, from Bing’s grounding queries report (step 4), and from the follow-ups customers ask in sales and support.

2.2 Lead every section with a direct, self-contained answer

Open each question-led section with 40 to 60 words that answer it completely. Microsoft describes the target as sentences that “make sense even when pulled out of context” and concise answers of one to two sentences that directly address the question. A sentence that depends on the paragraph above it does not get quoted; a competitor’s cleaner sentence does.

2.3 Replace generalities with quotes, statistics and citations

Research shows (the KDD 2024 GEO benchmark) that rewriting a source to add relevant quotations raised its share of the generated answer by about 41 percent on the paper’s main metric, adding statistics by about 33 percent and adding citations by about 28 percent, while keyword stuffing lowered it by about 9 percent. The engine was GPT-3.5 over Google’s top five results, the sources were already in the candidate set, and the gains went mostly to lower-ranked sources, so treat these as direction rather than a forecast. Microsoft’s AI Performance guidance says the same thing in prose: “examples, data, and cited sources help build trust when content is reused”.

In practice: every claim gets a number, a named source and a date; the source is linked and primary; and at least one passage on the page is something another writer would quote (a definition, an original figure, a comparison).

2.4 Be non-commodity

Google says AI search users ask longer, more specific and follow-up questions, and that the content to create is “non-commodity”, in its May 2025 Search Central guidance. A page that says what every other page says is interchangeable to a model that has read all of them. Original data from your own operations, a tool, a worked example with real numbers, or a documented test is what a summary cannot substitute for.

2.5 Make the author and the entity unambiguous

Answer engines attribute. Name the author and the organization on the page, keep the description of what you are identical across your About page, your social profiles and third-party listings, and keep structured data consistent with the visible text. Google’s guidance is explicit that markup content must also be visible on the page. There is no measured citation lift from schema itself (Ahrefs, May 2026), so treat it as hygiene.

Our ChatGPT SEO and Perplexity SEO posts apply this step to each engine’s habits.

Step 3: Be mentioned elsewhere

AI systems weigh what the rest of the web says about you, and the only large-scale measurement of that points at mentions rather than links.

Research shows (Ahrefs, 75,000 brands, May 2025) that branded web mentions had a Spearman correlation of 0.664 with a brand’s visibility in AI Overviews, the strongest factor tested. Branded anchors came in at 0.527, branded search volume at 0.392, Domain Rating at 0.326, referring domains at 0.295 and backlinks at 0.218. The authors say, in their own words, that correlation is not causation; the sample was limited to domains with a Domain Rating above 40; and the measurement was AI Overviews only. No equivalent study exists for ChatGPT or Perplexity, so do not assume the coefficient transfers.

What that justifies doing:

1. Get reviewed and compared. Product comparisons, “best of” lists on sites with editorial standards, and review platforms are the pages an engine retrieves for commercial questions. Your presence there is a mention with context.
2. Show up where the question is discussed. Community threads, industry publications and expert newsletters are heavily retrieved for “what is the best” and “how do I” questions. Participation has to be real; a drive-by link is the thing those communities delete.
3. Be quotable, and be quoted. Publish the figure, the definition or the test that other writers need, and they will cite you. That is earned media, and it produces the mention and the link together.
4. Keep the facts identical everywhere. The founding year, the price, the feature list and the one-sentence description should match across your site, profiles and listings. An engine reconciling conflicting facts about you has a reason to prefer a competitor it can describe confidently.

What it does not justify: buying mentions on listicles or directories. There is no evidence a bought mention moves a model, low-quality pages are the least likely to be retrieved, and Google says creating links “primarily for the purpose of manipulating search rankings” is link spam under its spam policies, which govern the index every AI feature reads.

Step 4: Measure

Measure AI visibility in four places, none of which is complete on its own. Google reports impressions, Bing reports citations, GA4 reports referrals, and a prompt panel you run yourself fills the gaps for the engines that report nothing.

4.1 Google Search Console: the generative AI performance report

Google says the Search generative AI performance reports, announced in June 2026 and available to all sites since August 31, 2026, show how often your URLs appeared in AI Overviews and AI Mode, by page, country, device and date down to the hour. The report is impressions only. Clicks from those features remain inside the overall Performance report under the “Web” search type, blended with ordinary results. Use the report to learn which pages appear in AI features and how that changes after you edit them; do not expect it to show queries.

4.2 Bing Webmaster Tools: the AI Performance report

Microsoft’s AI Performance report, in public preview since February 2026, covers Copilot, Bing’s AI summaries and partner integrations. It shows total citations, average cited pages per day, page-level citation counts and a sample of the grounding queries the AI used to retrieve each page. The June 2026 update added intent and topic groupings, a period comparison, and Citation Share, “the percentage of citations attributed to your site out of all citations shown across all sites” for the same grounding query. Microsoft states that citations are not traffic, clicks or engagement. It is still the only report on any platform that shows you the question an AI actually asked your page, which makes it the best input to step 2.1.

4.3 A prompt panel for ChatGPT, Perplexity, Claude and AI Mode

Nothing reports citations from these engines to publishers. The repeatable substitute is a panel you run by hand or with a tool.

1. Write 20 to 50 prompts in the words customers use, split across informational, comparison and purchase intents.
2. Run each prompt in each engine on the same day each month, logged out or in a fresh session, and record the cited domains, whether yours appears, where it appears, and the sentence attached to it.
3. Compute your citation rate per engine (prompts citing you divided by prompts run) and your AI share of voice (your citations divided by all citations for those prompts).
4. Keep the prompt list fixed for at least a quarter. Changing prompts and content at the same time tells you nothing.

A minimal log looks like this.

Prompt	Engine	Date	Cited domains	Ours cited?	Position	Sentence attached to our citation
best free seo audit tool	ChatGPT	2026-11-03	4	Yes	2	“…runs 75 checks including AI crawler access…”
best free seo audit tool	Perplexity	2026-11-03	6	No

The AI visibility checker covers the readiness side of this for a single URL: which bots can reach it, whether it is indexed and snippet-eligible, and whether the content is extractable without JavaScript. The outcome side is the panel.

4.4 GA4: AI referral traffic

Visits from chatgpt.com, perplexity.ai, copilot.microsoft.com, claude.ai and gemini.google.com arrive in GA4 as referrals. Create a custom channel group with a rule matching those hostnames (and their subdomains) so AI referrals appear as one line in acquisition reports, and attach conversions to it. Two caveats. Clicks from AI Overviews and AI Mode arrive as google / organic and cannot be separated from ordinary search clicks in GA4. And Google says to measure “the full value of your traffic”, meaning conversions rather than sessions; our view is that this is fair, because AI referrals are fewer and better informed, and session counts will make the channel look smaller than it is.

Vendor claim: Microsoft Advertising reported that AI referrals to top websites rose 357 percent year over year in June 2025 to 1.13 billion visits. Treat that as Microsoft’s measurement of a trend, not a benchmark for your site.

Myths about AI visibility

• “You need an llms.txt file.” Google says Search does not use it, and no other engine documents reading it. It is harmless and optional; our llms.txt guide explains the narrow cases where it is still worth publishing.
• “Schema markup gets you cited.” Research shows (Ahrefs, May 2026, 1,885 pages against matched controls) no meaningful change in citations on AI Overviews, AI Mode or ChatGPT after adding JSON-LD, with the caveat that every page in the sample was already heavily cited. Keep markup accurate; do not buy it as a citation tactic.
• “Blocking GPTBot hides you from ChatGPT.” It keeps you out of training. OAI-SearchBot is the switch for ChatGPT search, and OpenAI documents the two as independent.
• “Perplexity ignores robots.txt, so there is nothing to do.” PerplexityBot, the indexing crawler, is the one to allow. Perplexity-User, the user-triggered fetcher, is what ignores robots.txt, and it fetches a page only when someone asks about it.
• “AI Overviews use a different crawler or need special markup.” They use Googlebot, the Search index and the same snippet controls. Google says there is “no special schema.org structured data that you need to add”.
• “An AI visibility score tells you where you stand.” Any single score, including ours, is a readiness measure of a page, not a measurement of citations. Citations come from the two publisher reports and your panel.
• “AI Overviews do not cost clicks” and “AI Overviews destroy traffic.” Both overstate. Research shows (Ahrefs, February 2026) a 58 percent lower position-one click-through rate with an AI Overview present, on desktop, against a forecast baseline; Pew’s 2025 panel found 8 percent versus 15 percent click rates; Google says the remaining clicks are higher quality. The methods differ, so the numbers do. Measure your own.

What to do next

Work the steps in order and do not skip the first. Today: check robots.txt for the five search bots, confirm indexing in both Google and Bing, and remove any snippet restriction you did not choose. This week: set up the Search Console generative AI report, the Bing AI Performance report, a GA4 channel for AI referrals and a 20-prompt panel, so there is a baseline before anything changes. This month: rewrite your three most important pages so every section leads with a self-contained answer backed by a dated, sourced number, and move anything hidden in tabs into the HTML. This quarter: earn mentions where your category is discussed and keep your facts identical everywhere. Then read the panel. For engine-specific detail, continue with how to get cited by ChatGPT and AI Overviews optimization.

Frequently asked questions

How do I get cited by ChatGPT?

Allow OAI-SearchBot in robots.txt (blocking GPTBot does not matter for search), make sure the page is indexed and readable as plain HTML, write sections that answer a question directly with specific sourced facts, and be mentioned on third-party sites ChatGPT already trusts. There is no publisher report; measure with a prompt panel and GA4 referrals from chatgpt.com.

How do I rank in AI Overviews?

You do not rank in them; you are cited by them. Google says a page needs only to be indexed and eligible to show with a snippet, with no special markup or files. Both AI Overviews and AI Mode use normal Googlebot and may fan a question out into several sub-queries, so pages that answer the sub-questions specifically get picked up.

Does Perplexity respect robots.txt?

PerplexityBot, the indexing crawler, is the one Perplexity recommends allowing. Perplexity-User, which fetches pages when a person asks a question, generally ignores robots.txt according to Perplexity's own documentation, because the request was user-initiated.

Do brand mentions improve AI visibility?

They correlate with it. Ahrefs' study of 75,000 brands found a 0.664 Spearman correlation between branded web mentions and AI Overview visibility, far above backlinks at 0.218. The authors state that correlation is not causation, so earn mentions rather than buy them.

Which reports show AI visibility?

Google Search Console's generative AI performance report (impressions in AI Overviews and AI Mode by page, country, device and date) and Bing Webmaster Tools' AI Performance report (citations, cited pages, grounding queries and Citation Share for Copilot). ChatGPT, Perplexity and Claude report nothing; use a prompt panel and GA4 referrals.

Is an llms.txt file needed for AI visibility?

No. Google says it does not use llms.txt, and OpenAI, Anthropic and Perplexity do not document any use of it. It is harmless to publish but does nothing for inclusion or citation.