How to get your company cited by ChatGPT, Perplexity and AI Overviews

What do AI search tools actually read?

Each AI search tool answers from a search index built by a crawler, plus live page fetches when a user asks something specific. If that crawler cannot reach your page, or your text only appears after JavaScript runs, the tool has nothing of yours to cite.

The engines are not one system. They use different crawlers, and each one is controlled separately in robots.txt.

Google's position is the simplest. Its AI features documentation (opens in a new tab) says "there are no additional requirements to appear in AI Overviews or AI Mode." A page has to be indexed and eligible to show a snippet. That is the whole entry ticket.

The other engines add one technical trap. Vercel's study of AI crawlers (opens in a new tab) (December 2024) found that none of the major AI crawlers render JavaScript. OpenAI's and Anthropic's crawlers fetch script files without running them. Googlebot and Applebot are the exceptions. If your headings and copy are injected by client-side code, ChatGPT and Claude see an empty page. We render every word of this site on the server for that reason, and the 3D work sits behind the text as decoration.

Rankings still matter, though less directly than they did. Ahrefs found that 76 percent of AI Overview citations came from top-10 pages in July 2025. A follow-up across 863,000 keywords, reported in March 2026 (opens in a new tab), put that share at 38 percent, with about 31 percent from positions 11 to 100 and 31 percent from beyond 100. Ahrefs says better parsing makes the two datasets not directly comparable, and suggests query fan-out as one cause. The practical reading: cover the sub-questions a buyer's query splits into, not only the head term.

Which crawlers should you allow?

Allow every search crawler and user-fetch agent you want citations from. Training crawlers are a separate decision, and blocking them does not remove you from search answers.

Sources: OpenAI crawler documentation (opens in a new tab), Anthropic crawler documentation (opens in a new tab), Perplexity crawler documentation (opens in a new tab) and Google's list of common crawlers (opens in a new tab). OpenAI notes that robots.txt changes for OAI-SearchBot take about 24 hours to apply.

Our default for a company that sells expertise is to allow all of them, training bots included. Being known to future models costs you nothing if your content is marketing. A publisher whose content is the product should weigh training access differently.

Two traps catch people. First, a crawler obeys only the most specific robots.txt group that names it. Add a group for GPTBot and the disallow lines under the catch-all group no longer apply to GPTBot, so private paths need repeating in every named group. Second, bot protection at your CDN or firewall can block AI crawlers before robots.txt is ever read. Check your server logs for each user agent, not just the robots file.

How do you write a page that AI answers can quote?

Put a real question in the heading, answer it in 1 to 3 sentences directly below, then give the detail, the table and the source. A passage that makes sense when lifted out on its own is a passage an AI answer can use.

Microsoft's Bing team says it plainly in its October 2025 guidance (opens in a new tab): "Direct questions with clear answers mirror the way people search." The same post warns that "AI systems may not render hidden content," so answers that only load through JavaScript after a click are the most likely to be missed. Answers already in the HTML, even inside a closed accordion, are at least there for a crawler to read, which is why this site keeps them in the page. Google's AI optimization guide (opens in a new tab) pushes in the same direction from another angle: it asks for content that offers "unique expert or experienced takes that go beyond common knowledge."

The best evidence on writing tactics is still the original GEO paper (opens in a new tab) (Aggarwal and others, KDD 2024). Its lab benchmark measured how much of a generated answer drew on a source after each rewrite:

Two caveats. This is a 2023 to 2024 lab benchmark, not a production engine, so treat it as direction rather than a forecast. And the gains were largest for lower-ranked pages: adding citations lifted the fifth-ranked source by 115 percent in the paper's per-rank analysis. That is good news for a new site.

Our quotable page checklist, which we apply to every page we write:

1. One topic per URL, with the title, H1 and meta description saying the same thing.
2. A 40 to 80 word answer at the top that would still make sense pasted on its own.
3. Question headings in the buyer's words, each opening with a direct answer.
4. A one-sentence definition of the main term, linked to a glossary entry where one exists.
5. A table wherever the content is a real comparison.
6. Every number with a linked, dated source, or labeled as our own estimate.
7. A visible updated date that changes only when the content does.
8. All of it in the server-rendered HTML, not in images, PDFs, canvas or click-to-open panels.

Why does a consistent company name matter?

AI answers name companies, so the engine has to be sure which company a page is about. If your name, description and founders differ between your site, LinkedIn and directories, the model has to guess, and a guess rarely ends in a citation.

Bing's Webmaster Guidelines added grounding guidance in February 2026 that says "entity names should be clear and consistent" (Search Engine Journal summary (opens in a new tab)). In practice, fix these in one sitting:

• One form of the company name. If people shorten it, declare the short form as an alternate name in your Organization schema instead of using both at random.
• One sentence that says what you do and for whom, reused word for word on your site, LinkedIn and every directory profile.
• The same logo file, founder names and primary URL everywhere.
• Organization schema with links only to profiles that exist. Our entity SEO entry covers the fields.

This is dull work. It is also the cheapest fix in this article, and it compounds, because every new mention reinforces the same description.

How do you earn mentions on pages AI answers already cite?

Run your buyers' questions through the assistants, log which URLs they cite, and earn an accurate place on those pages. Google's guide is blunt about the shortcut: "seeking inauthentic 'mentions' across the web isn't as helpful as it might seem."

The routine we run:

1. Write 30 to 50 prompts your buyers would actually type. For example: "best AI automation agency for a 50-person law firm" or "how much does a RAG system cost".
2. Run them monthly in ChatGPT, Perplexity, Gemini, Google AI Mode, Claude and Copilot, and log every cited URL.
3. Group the cited pages: directories and review sites, comparison articles, community threads, vendor partner listings and competitors' pages.
4. Pick the honest route into each group. Complete and correct your directory profiles. Send comparison authors original data or a factual correction. Answer in communities as a named person who discloses the company. Join partner programs you genuinely qualify for.

Skip anything you would not want a buyer to find out about later. Bought reviews and planted posts tend to surface, and Bing's guidelines now have a full section on "Prompt Injection and AI Manipulation."

Do structured data and llms.txt help you get cited?

Neither one switches citations on. Structured data helps machines confirm who you are and what a page is. Google says you need no special schema, files or markup for its AI features.

Google's AI optimization guide says "you don't need to create new machine readable files, AI text files, markup, or Markdown to appear in Google Search." The FAQ rich result also stopped appearing in Google Search (opens in a new tab) on May 7, 2026, so FAQ markup no longer earns a visual result there.

We still ship schema markup (Organization site-wide, Article on articles, Service on service pages, BreadcrumbList on inner pages), because it states identity in a form Bing, Google and LLM parsers can read, and it costs little when generated from the same content as the visible page. The rule is that markup must match what a visitor can see. An llms.txt file is in the same category: Google Search ignores it, some agents read it, and it is worth an hour if your pages are generated from one content source so the two never drift apart.

How do you measure AI citations?

Use 4 sources together, because none of them is complete on its own: Search Console for Google impressions, Bing Webmaster Tools for citations, analytics for referral visits and a prompt panel you run yourself.

Google rolled the generative AI report out (opens in a new tab) to all websites as of August 31, 2026. Microsoft describes the AI Performance dashboard (opens in a new tab) as covering citations, cited pages and grounding queries, with a mapping from each query to the page it cited. Those grounding queries are the most useful content roadmap we know of: they show the phrasing an AI used to retrieve your page.

Rules for the prompt panel, so the numbers mean something month to month: keep the prompts fixed, run them on the same day each month, use signed-out or fresh sessions where the tool allows, run each prompt twice, and log mention, citation and URL as separate columns. A brand named without a link is worth knowing about, but it is a different result from a citation.

What are we tracking on our own site?

The Method Crew launched in 2026, so we have no citation results worth publishing yet. We will not quote any until we have 3 months of panel data, and when we do, they will go here with dates.

The setup is the one described above: a fixed panel of buyer prompts about AI consulting costs, automation platforms and AI search, run monthly across 6 assistants, plus Search Console and Bing Webmaster Tools data. It is the same reporting included in our SEO and GEO retainer. The AI search content engine blueprint shows how the same panel feeds a content program: questions in, reviewed pages out, citations tracked.

What does not work, and when is this the wrong spend?

Hidden instructions to AI systems, a page for every variant of a query and fake freshness all fail or get penalized. Citation work is also the wrong spend when your buyers never research in assistants or your site has basic indexing problems.

What fails:

• Hidden text or instructions aimed at language models. Bing treats this as abuse.
• Publishing a page for every phrasing a fan-out might produce. Google's guide calls this scaled content abuse.
• Changing dates without changing content. Google's helpful content guidance (opens in a new tab) names it.
• Blocking OAI-SearchBot, PerplexityBot or Claude-SearchBot to protect content you want cited.
• Restrictive snippet controls such as nosnippet or a low max-snippet on pages you want quoted.

When to spend elsewhere: if your work comes almost entirely from referrals and repeat clients, citation work will not move revenue. If Google has not indexed your key pages, fix that first through an SEO and GEO audit. If you need leads within 30 days, paid search is faster. And if you have nothing original to say about your field, no amount of formatting will make a page worth quoting.

Nobody can guarantee a citation, and Bing's own guidelines say so about GEO. What we sell is the work that makes one likely: an answer engine optimization sprint at $6,000 to $15,000, an audit at $3,500 to $8,000, or a monthly retainer at $3,000 to $8,000 with the reporting above. For the difference between the labels, read GEO vs AEO vs SEO.