How to Show Up in ChatGPT and AI Search

The short answer

To appear in ChatGPT you need three things, in this order: be indexed in Bing (ChatGPT’s live search reads Bing’s index, not Google’s), be described consistently across the web so the model can tell who you are, and be mentioned on sources other than your own site. Publishing more pages on your own domain is the least effective of the three, and we have our own data showing why.

First: the prerequisite almost nobody checks

ChatGPT’s web search retrieves from Bing’s index, supplemented by OpenAI’s own crawler. The practical consequence is blunt: if Bing has not indexed your page, ChatGPT cannot cite it, no matter how well it ranks on Google. Most businesses have never opened Bing Webmaster Tools.

An admission, because it makes the point better than a client example would. We tell everyone this is the most-missed prerequisite in AI visibility. Last week we finally audited our own: sempite.com had 40 of its 212 pages in Bing’s index. We were giving correct advice we had not fully taken. If you do one thing from this guide, verify your site in Bing Webmaster Tools, submit your sitemap, and check how many pages are actually in there. It is free and takes ten minutes.

Ranking on Google does not carry over

This is the assumption that costs businesses the most time. We built a monthly study to test it: we take the businesses ranking in Google’s local top three and ask ChatGPT to recommend a business in that category and city. In the current run, 43% of the businesses Google ranks in its top three are never mentioned by ChatGPT at all. Being the best result on one system buys you nothing on the other. The full method and the month-over-month numbers are in the AI Visibility Index.

What actually predicts whether an AI names you

Ahrefs studied 75,000 brands to see which factors correlate with appearing in AI answers. Branded web mentions — people writing about you on sites you do not control — correlated at 0.664. Backlinks, the traditional SEO currency, managed 0.218. The strongest signals were all off-site.

One honest caveat, which the study’s own authors make and most summaries drop: correlation is not causation. Well-known brands naturally accumulate mentions and AI visibility at the same time, so brand strength is a plausible common cause for both. The finding does not prove that buying mentions buys citations. What it does tell you is where to look, and it points away from your own website.

Our own data, from the other direction

We tested the same idea by taking a category apart. Across 40 buying-intent queries in sports nutrition, we logged every source Google’s AI cited. Only 5.1% of citations went to brand-owned sites. The brands being discussed were almost never the sources being quoted — publishers were. That is the Ahrefs correlation seen from the other end, and it reframes what your own website is for: not to be the citation, but to make you legible enough that the sources which do get cited describe you correctly. Method and full source log in that study.

The gatekeeper problem

Citations are not spread evenly. In the same category, three publishers carried 77% of the AI answers — and all three publish the criteria they score products on. That is unusually good news: where a handful of sources dominate a niche, the work is not vague brand-building, it is identifying those sources and meeting their published criteria. We wrote up who they were and what they actually ask for in the gatekeepers study.

Your category will have its own three. Ask ChatGPT ten buying-intent questions in your niche and record which domains it cites. That list, not a generic directory checklist, is your target.

What ChatGPT does when it has never heard of you

Worth knowing, because “invisible” has two different failure modes. We asked three AI systems about the same 13 performers on the same day. For the seven who were real but not famous, ChatGPT declined all seven rather than guess. Gemini invented a pageant title and a telenovela role. Google looked them up and got it right.

So the risk is not only being absent. It is being filled in. And the fix for both is the same: give the machines something specific, consistent and verifiable to retrieve. The detail is in Decline, Fabricate, or Retrieve.

Does llms.txt help?

Short answer: not yet, and not much. It costs nothing to publish a good one, but do not treat it as a lever. We audited 1,563 llms.txt files across the public directories — one in six was already broken. We opened the 1,318 that worked: 13% were completely empty, and the median contained just 30 links. We fetched 10,858 Shopify stores and found 98% serving a byte-identical auto-generated file that maps no content at all. And when we crossed our llms.txt data against who Google’s AI actually cites, the relationship ran backwards: only one of the ten most-cited publishers published an llms.txt at all.

Publish one if you want — our free generator and validator will make a correct one in a minute. Just do not expect it to move your visibility, and be suspicious of anyone selling it as the answer. The four studies are here.

The checklist that actually moves it

1. Verify in Bing Webmaster Tools and submit your sitemap. Check your indexed page count against your sitemap count. Fix the gap before anything else.
2. Make one page state plainly who you are, what you do, and who you serve. Answer-first: the claim in the first sentence, evidence after. Models lift self-contained statements and skip buried ones.
3. Make every profile agree. Same name, same description, same category across your site, your Google Business Profile, LinkedIn, and every directory. Contradictions read as unreliability to a system cross-checking sources.
4. Find your three gatekeepers. Ask ten buying-intent questions in your niche, record the cited domains, and work on being present and accurate on those specific sites.
5. Earn mentions you do not control. Industry directories, local press, podcasts, review sites, expert quotes. This is slow and it is the thing the data keeps pointing at.
6. Publish something worth citing. Original data is the cheapest way onto pages you do not own — one study can earn mentions across a dozen sites that would never link to a services page.
7. Add FAQ and Organization schema. Not magic, but it makes extraction reliable.

How to measure whether it is working

Pick fifteen questions a real customer would ask — the ones ending in a hiring decision, not your brand name. Ask each one monthly in ChatGPT, Perplexity and Google’s AI Mode. Record three things every time: were you named, was what it said correct, and who was named instead. That last column is the useful one; it is a live list of who is beating you and on which source.

Our free AI Citation Checker runs the first pass, and the AI Readiness Score covers the technical half — whether engines can read you at all. If you would rather wire it into your own reporting, the same checks run through the AI Search Visibility API.

How long it takes

Faster than classic SEO for the retrieval half, slower for the memory half. Perplexity and Google’s AI read live pages, so structural fixes can show up within weeks. ChatGPT blends live retrieval with training data, so the part that depends on the model having absorbed a consensus about you moves on the order of months and follows model updates you do not control. Fix the Bing gap and the consistency problems first — those are the fast ones — then treat mentions as a standing habit rather than a campaign.

If you want the wider playbook across every engine rather than ChatGPT specifically, that is the AI Search Visibility guide. If you want to know where you currently stand, the free visibility check takes about sixty seconds.