How ChatGPT Describes Your Brand, and Why

Every description ChatGPT gives of your company comes from one of two places, or a blend of both. The first is what the model learned in training. The second is what it fetches from the web while answering. They break in different ways, so the fix depends on which one produced the answer.

This split matters because an outdated answer and an invented one need different responses. Our breakdown of a hallucination versus an outdated answer covers that diagnosis in more depth.

When ChatGPT answers without searching, it relies on what the model absorbed in training. That knowledge stops at a cutoff date. OpenAI's model documentation, as verified in August 2026, lists a February 16, 2026 knowledge cutoff for the GPT-5.6 family and August 31, 2025 for the Instant model behind the chat-latest alias. OpenAI also warns these snapshots are regularly updated, so check the current values before you rely on them.

Two things follow. First, "ChatGPT" is not one knowledge state. Depending on which model handles a request, its picture of your company can differ by about five and a half months. Second, anything you launched, renamed, repriced or retired after the cutoff does not exist for that model. If you dropped a feature last year, a training-data answer may still describe it. See what to do when AI describes features you dropped.

Nobody outside OpenAI can edit this layer. GPTBot crawls content that may be used to train future models, and OpenAI says disallowing it signals your content should not be used for training. OpenAI makes no claim that allowing GPTBot improves how you appear in answers, and no published study measures that effect in either direction. Any change you make reaches this layer, if at all, only when a future model is trained.

When ChatGPT searches, it builds its description from pages it retrieves at answer time. OpenAI's help centre says ChatGPT search typically rewrites your question into one or more targeted queries and sends them to search providers. Your pages enter that pool through OpenAI's search crawler, OAI-SearchBot.

OpenAI's crawler documentation states the consequence plainly: sites opted out of OAI-SearchBot will not be shown in ChatGPT search answers, though they can still appear as navigational links. It also says each crawler setting is independent of the others. You can block training and still allow search:

User-agent: OAI-SearchBot
Allow: /

User-agent: GPTBot
Disallow: /

robots.txt is only half the gate. OpenAI's help centre says eligibility also requires your host or CDN to allow traffic from OpenAI's published searchbot IP addresses. A firewall that returns 403 to OAI-SearchBot removes you as completely as a robots.txt block, and no dashboard will tell you. Cloudflare's new defaults from September 15, 2026 block the Training and Agent crawler categories on ad-displaying pages for new domains, new sites on existing accounts and free-tier customers, while Search stays allowed. Check your own settings rather than assuming.

Access also means the crawler can read the content. Vercel's crawler study found ChatGPT's crawlers fetch JavaScript files but do not execute them, and independent re-tests as of June 2026 found the same for OAI-SearchBot and ChatGPT-User. Text that only appears after JavaScript runs is invisible to them.

One more bot: ChatGPT-User fetches pages when a user's action triggers it. OpenAI says robots.txt rules may not apply to it, and it is not used to decide what appears in search.

You can't see inside the model, but answers leave signals. Run this sequence for each buyer question that matters to you.

1. Look for sources. A search-backed answer usually shows cited links. No citations means you can't confirm search ran, and the answer may come from training data alone.
2. Ask about a fact that changed recently. Pick something that changed after the model's cutoff: a new price, a renamed product, a new office. If the answer gives the old version with no sources, training data produced it.
3. Ask again with search requested. Ask ChatGPT to search the web for the same question. If the description changes, the gap between the two versions shows how stale the training-data answer is.
4. Open the cited pages. Check whether the sentence ChatGPT gave actually appears on the page it cited. OpenAI's developer documentation notes that the number of sources consulted is often greater than the number cited, so citations are not a full record of what shaped the answer.
5. Test logged out, with memory off. OpenAI says ChatGPT may use saved memories when rewriting a search query, so a logged-in test reflects your own history. Our comparison of logged-in and logged-out AI testing explains why the results differ.

Don't expect a clean split every time. ChatGPT decides per query whether to search. Published estimates of how often it searches range from about 18% of conversations (Profound) to 80.5% of prompts (cloro), and the two measure different things on different populations. Treat any single answer as one sample. To judge whether what you see is correct, follow how to tell whether ChatGPT is accurate about your company.

In your control

• Whether robots.txt allows OAI-SearchBot.
• Whether your CDN or firewall lets it through.
• Whether your key facts are in the HTML before JavaScript runs.
• Whether those facts are stated plainly and the same way across your site. A specific fact on a fetchable page (a price, a founding year, a one-line description) gives the engine a sentence it can quote.

Partly in your control

• What third-party sites say about you. When ChatGPT cites directories, review sites or press instead of your site, being named accurately on those pages is associated with appearing in the answer. That is an association, not a promised effect.

Not in your control

• The training data and the knowledge cutoff.
• Whether ChatGPT decides to search for a given question.
• How it rewrites the question into search queries.
• Which retrieved pages it chooses to cite, and how it phrases the result.

The last list is long, and that's the honest picture. A July 2026 critical survey of 45 GEO studies found no technique with a stable, longitudinal, cross-platform causal effect on discoverability. What holds up is access: removing the blocks that keep your pages out of the search path.

Work from the gate inward. Each step only pays off once the one before it passes.

1. Record what ChatGPT says today. Save answers to five to ten real buyer questions, logged out, with and without search requested. Note which answers carry sources. Our guide to finding out what AI says about your company lays out how to build this baseline.
2. Check robots.txt per bot. Confirm OAI-SearchBot is allowed. Decide on GPTBot separately, as a training-use question.
3. Check the network layer. Request a page as OAI-SearchBot and confirm it gets a 200, not a 403 or a challenge page. The AI Crawler Checker requests a page as each AI crawler and shows which ones your firewall turns away.
4. Check rendering. Fetch the raw HTML with JavaScript off. Confirm your pricing, product names and one-line description are there.
5. Make your own facts agree. Your About, pricing and product pages should state the same facts in the same words. See what your About page should say for the facts to include.
6. Correct third-party pages you can influence. Update directory listings and profiles that ChatGPT cites for your category.
7. Re-test on a schedule. Answers vary by query and over time. Monitoring what ChatGPT says about your brand covers cadence and what to log.

If you want a quick starting point, the free checker on this site gives a snapshot of whether AI engines name you for up to three buyer questions. Treat it as a first look, not continuous tracking.

• Blocking every AI bot with one rule. A blanket block on AI crawlers silently removes ChatGPT search eligibility. Set rules per bot.
• Allowing GPTBot "so ChatGPT can find you". GPTBot governs training only. Search uses OAI-SearchBot.
• Treating one logged-in answer as the truth. Memory, model routing and the per-query search decision all change the result.
• Expecting a page update to fix a training-data answer. An update can reach search answers after recrawl. It reaches training-data answers only through a future model.
• Adding schema, an llms.txt file, or an "AI-friendly" rewrite. Ahrefs' matched test of 1,885 pages found adding schema produced no major uplift in citations on any platform. No engine documents reading llms.txt, and Google says you don't need to write in a special way for AI search.
• Assuming Bing rankings carry over. In Otterly's 2026 experiment, two sites ranked in Bing with ChatGPT's crawlers visiting, yet earned effectively zero ChatGPT citations.
• Hiding text or instructions aimed at AI crawlers. Prompt injection and cloaking exist, and research shows they can sway some systems once the content is retrieved. They are manipulation, they risk removal from search, and they are not a legitimate fix.

Does blocking GPTBot remove my brand from ChatGPT? No. OpenAI says each crawler setting is independent. GPTBot governs whether your content may be used to train future models. OAI-SearchBot governs whether you can appear in ChatGPT search answers. Blocking GPTBot leaves search eligibility untouched, as long as OAI-SearchBot is allowed and your firewall lets it through. Whether blocking training changes what future models recall about you is unmeasured by anyone.

How long until ChatGPT reflects a change on my website? It depends on the source. Search-backed answers can reflect the change once OAI-SearchBot has recrawled the page, though OpenAI publishes no timing. Training-data answers won't reflect it until OpenAI ships a model with a later knowledge cutoff, which can take many months. Until then, the same question can return the new fact with search and the old fact without it.

Why does ChatGPT describe my company differently for different people? Several things vary per request. ChatGPT decides per query whether to search. It rewrites the question into its own search queries. It may use saved memories when rewriting, and different users can be routed to different models with different cutoffs. Two people asking the same question can therefore get answers from different sources. Test logged out, with memory off, and repeat each question.

Does ChatGPT use Bing to describe my brand? Sometimes. OpenAI's help centre says ChatGPT search sometimes partners with other search providers, but it doesn't quantify how much any provider contributes. Ranking in Bing is not a reliable route in. In Otterly's 2026 experiment, sites that ranked in Bing, with ChatGPT's crawlers visiting, received effectively zero ChatGPT citations. Make sure OAI-SearchBot can reach you directly.

If ChatGPT cites my page, did it use my page? Not necessarily, and not only. OpenAI's developer documentation notes that the number of sources consulted is often greater than the number cited. Mechanistic research on how language models cite has shown citation output can be repaired without changing the sources, which suggests citation is partly separate from influence. Read the cited page and confirm the claim is actually there.