How to Rank in ChatGPT: A Measured Answer
How to rank in ChatGPT: the three requirements, and what 39 measured answer runs show about who AI engines actually name in B2B categories today.
Your company published forty pages this year. Ask ChatGPT who does what you do, and it names somebody else — or nobody at all. Meanwhile the advice on offer assumes you are losing a crowded race. In the B2B categories we measure, you are not losing a race. Nobody has entered it.
To rank in ChatGPT, your site needs three things: technical access (AI crawlers must be allowed and able to reach your content), citable substrate (structured, factual content an answer engine can extract and attribute), and off-site presence across sources ChatGPT already trusts. Most sites we have surveyed have never even addressed step one — not blocked, just undecided and invisible by default.
What "ranking" means when ChatGPT has no ranked list
ChatGPT does not maintain a position-ordered index the way Google does. When it answers a buying question, it composes a fresh answer — often from a live search — and the brands it names change from run to run. Ahrefs, citing SparkToro research, reports that if you ask ChatGPT the same question a hundred times, the chance that any two responses contain the same list of brands is under one percent (ahrefs.com/blog/how-to-rank-on-chatgpt, read 2026-08-27).
So "ranking" in ChatGPT is not a position. It is a probability: how often you get named when a buyer asks a question you should be the answer to. That has two practical consequences. First, a single check tells you almost nothing — you have to measure repeatedly and count. Second, the goal is not to "win the keyword" but to become the entity the engine reaches for most often. Everything below serves that probability.
How to rank in ChatGPT: the three requirements
Technical access — robots.txt, OAI-SearchBot, and why Bing matters
Before any content tactic, the machines have to be able to read you. ChatGPT's ecosystem fetches your site with named crawlers — OAI-SearchBot for search results, GPTBot for training, ChatGPT-User for live page visits. OpenAI documents each one, with its exact user-agent token, at developers.openai.com/api/docs/bots (read 2026-08-27). If your robots.txt or your CDN blocks them, no amount of content work matters.
The second, less obvious dependency is Bing. As Omnius puts it: "Most people aren't talking about this yet, but ChatGPT relies on the same ranking signals as Microsoft Bing." (omnius.so/blog/how-to-rank-on-chatgpt, read 2026-08-27.) A site Bing has never crawled is structurally absent from a large part of ChatGPT's grounding, whatever Google thinks of it. Registering in Bing Webmaster Tools and submitting a sitemap is unglamorous and load-bearing.
And in practice, the technical floor is failed far more often than the advice industry assumes. In August 2026 we surveyed the robots.txt of 23 large German online shops: 19 of them named no AI crawler at all, 3 explicitly allowed at least one, and 1 blocked GPTBot only (our own curl survey, 2026-08-15, method and raw results documented). One major retailer served HTTP 404 to OpenAI crawler user-agents while a browser got 200 — measured at the user-agent level, so real bot IPs may behave differently, but that is exactly the kind of silent failure a five-minute check surfaces.
One more dependency worth naming: what an AI crawler receives is your raw HTML. In our own site crawls we check every page's raw HTML for exactly this reason — a client-rendered React or Vue shell can look perfect in a browser while the HTML the crawler actually receives is missing the visible content entirely.
Citable substrate — content an engine can extract and attribute
Answer engines do not reward cleverness; they reward extractability. The pages that get cited share a shape: the question a buyer actually asks appears verbatim as a heading, the answer follows in the first sentences after it, facts are dated, and claims are specific enough to quote. A 3,000-word essay that circles its subject gives an engine nothing to lift.
This is where answer engine optimization genuinely diverges from classic SEO — and why "llm seo" has become its own discipline. Google can rank a page that answers the question eventually. An answer engine composing a response needs the answer in liftable form, now, with a source it can name. Write for the extract, not the dwell time.
Off-site presence — the sources the engines already trust
In our measurements, the answers to German B2B buying questions are built from a substrate of small, specific sites — agency pages, an independent directory profiling 192 German AI providers (deutschlandki.de, read 2026-08-27), vendor documentation — not from big media. That is worth knowing for two reasons. It means the bar to becoming a cited source is lower than it looks. And it means classic digital-PR instincts (chase the biggest outlet) point at the wrong targets: the engines are already citing sites you could realistically be listed on, mentioned by, or compared in.
What we measured: 39 answer runs, no consistently named provider
Here is the part none of the ranking guides carry: an actual, dated measurement of who ChatGPT names in a real B2B category.
Our method: weekly measurement, five answer runs per query per engine (ChatGPT with search, and Perplexity), against four fixed buyer-intent queries — AI automation for the German Mittelstand, shop visibility to AI shopping agents, AI-visibility audits, and AI implementation for dental practices. On 2026-08-27, across 39 measured answer runs (one additional run failed and is excluded), the engines named plenty of sites — but no provider was named consistently across the runs, and AILoopwise was named in none of them. We publish that number because a measurement published only when it flatters you stops being a measurement.
The reading that matters for you: in this category — and, we suspect, in most B2B service categories — there is no incumbent to displace. The engines answer with whatever specific, extractable sources they happen to find. "How to rank in ChatGPT" is, today, mostly a question of showing up qualified: technically readable, extractably written, present on the handful of sources the engines already cite. That is a vacancy waiting for its first qualified entrant.
Two honest limits. Answers vary run to run, so any snapshot — including ours — is a sample, not a verdict; that is why we measure weekly rather than once. And anyone who promises you an AI ranking is selling a number they cannot control.
Answer engine optimization vs. traditional SEO — where they diverge
- No positions, only probabilities. You measure named-count over repeated runs, not a rank.
- A second index matters. Bing's crawl of your site feeds a large part of ChatGPT search's grounding; Google's does not.
- Freshness weighs more. Ahrefs measured ChatGPT's in-text references at roughly 393 days newer than organic Google results, and its citations at 458 days newer (ahrefs.com/blog/how-to-rank-on-chatgpt, read 2026-08-27).
- Extractability beats depth-for-its-own-sake. The liftable answer wins over the longest page.
- The technical floor is different. Allowing GPTBot and OAI-SearchBot, serving full HTML without JavaScript dependence, and being present in Bing are AEO requirements Google never forced on you.
What has not changed: real expertise, specific facts, and primary sources still beat generated filler — the engines are, if anything, better at ignoring filler than human readers are.
Brand visibility in AI search: what to build first when you are early
If your category has no AI-answer incumbent, order of operations matters more than volume:
- Clear the technical floor first. Check robots.txt against OpenAI's own bot list, fetch your key pages as those user-agents, register with Bing Webmaster Tools. Roughly 60 minutes of work — and 19 of the 23 large shops we surveyed had made no crawler decision at all.
- Give each real buyer question one answer-shaped page. The question verbatim as a heading, the answer in the first 40–60 words after it, dated facts beneath.
- Start measuring before you optimise. Ask the engines your buyer questions on a schedule, count who gets named, keep the raw answers. Without a baseline you cannot tell whether anything you change works. We explain what a meaningful check looks like in our guide to AI visibility checks.
- Get onto the substrate. Identify which sites the engines cite for your questions and earn a presence there — a directory entry, a comparison mention, a quoted fact.
If your business sells products rather than services, the mechanics differ — product visibility runs through feeds and merchant programs; see our walkthrough for getting products into ChatGPT.
This ordered list is our own working method as a Claude implementation partner doing AI-visibility work for B2B companies — the same method behind our AI-visibility service.
FAQ
Can you pay to rank in ChatGPT?
No. There is no advertising product that inserts your brand into ChatGPT's organic answers today. What you can buy is the work: technical readiness, extractable content, and presence on cited sources.
How long does it take to get named by ChatGPT?
Engines re-crawl and re-ground on their own schedules — changes typically need weeks, not days, before they can show up in answers. Anyone quoting a precise timeline is guessing.
Is ranking in ChatGPT the same as SEO?
They overlap — quality, specificity, and authority help both — but AEO adds its own requirements: AI-crawler access, Bing presence, extractable answer-shaped writing, and repeated measurement instead of position tracking.
Does Bing really matter for ChatGPT visibility?
Yes. ChatGPT's search layer draws on Bing's index and signals. A site absent from Bing is invisible to a large share of the grounding that produces answers.
How do I measure my AI visibility?
Ask the engines your real buyer questions repeatedly — we run five answer runs per query per engine, weekly — and count how often your brand is named. One-off free checkers give you a snapshot at best.
What is answer engine optimization (AEO)?
The discipline of making a brand appear in AI-composed answers — the three requirements above, plus repeated measurement. In Germany the same work usually runs under the label GEO (generative engine optimization) or KI-Sichtbarkeit.
Do llms.txt files help?
They are an emerging convention for describing your site to language models — and there is no measured evidence that publishing one improves how often engines name you. Treat it as cheap documentation, never as a visibility tactic; crawlability, extractable content and citations do the actual work.