Where does ChatGPT get its information about my business?

What sources does ChatGPT actually read about you?

Four layers, and only one of them is yours.

1. Your own website, fetched live. OpenAI documents a separate crawler for this: OAI-SearchBot is used to surface websites in search results in ChatGPT's search features. It is not the same bot as GPTBot, which collects training data, or ChatGPT-User, which fetches a page because someone asked ChatGPT to open it. Blocking the wrong one in robots.txt removes you from live answers.
2. Third party pages that mention you. Directories, industry association member lists, review platforms, supplier and partner pages, local news, government and council registers. This is the layer most businesses have never audited.
3. Community and reference sites. Forums, Q and A threads, Reddit, Wikipedia. Semrush studied 230,000 prompts across ChatGPT Search, Google AI Mode and Perplexity over 13 weeks from 14 July to 12 October 2025 and found Reddit and Wikipedia among the most-cited domains on all three.
4. The model's own training data. Whatever the model absorbed before its cutoff. You cannot see it, edit it or correct it. If ChatGPT knows your business without searching, this is where that came from.

The same Semrush study is also the reason not to bet a strategy on any one of these. It recorded ChatGPT's citation of Reddit falling from roughly 60% of responses in early August 2025 to around 10% by mid-September, a swing of about 50 points inside six weeks. Sources that carry a large share of citations today are not a fixed list.

Sources: OpenAI crawler documentation. Semrush, The Most-Cited Domains in AI, 230,000 prompts, 14 July to 12 October 2025, published 10 November 2025.

Why isn't a good website enough on its own?

Because in our own data, a good website barely predicts whether an engine names you.

We scanned 300 search-visible Australian businesses on 30 July 2026 and scored each site for AI readiness as well as counting what the engines said. Across the 137 businesses with complete records from both engines, the correlation between site readiness and the mean of the two per-engine visibility scores is Pearson r = 0.116 (n = 137). That is close to no relationship at all.

The gap is easier to see as two averages. Across all 300 sites, mean AI readiness was 74.9 out of 100. Across the same 300 businesses, mean AI visibility was 38.3 out of 100.

Sharper still, recomputed from our published dataset: 74 of the 300 sites scored a perfect 100 on site readiness. ChatGPT never named 36 of those 74 in any of the buyer questions we asked. Almost half of the technically flawless sites were invisible to ChatGPT anyway.

None of which makes on-site work optional. 13% of the 300 businesses fail the first bar by blocking AI crawlers outright in robots.txt, and a blocked site cannot be quoted at all. Readiness is necessary. It is nowhere near sufficient.

Source: Answerable, The State of AI Visibility in Australia 2026. 300 Australian businesses, six industries, scanned 30 July 2026. Raw data published as CSV and JSON under CC BY 4.0.

How do I find the sources a specific answer used?

Ask the question the way a customer would, then read every citation the answer carries.

1. Ask the buyer question, not your name. Best emergency plumber in Brunswick tells you something. Tell me about Acme Plumbing only tells you whether the model recognises you, which is a different and much easier test.
2. Open every citation and write down the domain. Not the page, the domain. You are building a list of the sites that get to speak about your category.
3. Repeat with three or four phrasings, on different days. One run is a sample of one. Answers move between runs, and a single result is not evidence of anything.
4. Do the same in Gemini. The two engines disagree constantly, so a source list built from one of them is half a list.
5. Check your own side. Look for OAI-SearchBot in your server logs, and confirm your robots.txt does not block it. A site that is never fetched is never quoted.

What comes out is a short list of domains that keep appearing for your category in your city. That is your actual target list, and it will not match a generic one from a blog post.

Which third party surfaces are worth the effort in Australia?

The ones your own answers already cite, and after that the registers a machine can verify you against.

An honest caveat first: our study did not measure the effect of individual platforms, so what follows is ordered by how checkable and durable each one is, not by an influence score we do not have.

• Your Google Business Profile. It feeds Google Search, and Gemini answers with Google Search grounding. Documented at every link, which is more than can be said for most recommendations in this field.
• Australian registers and professional bodies. ABN Lookup, ASIC's company register, AHPRA for registered health practitioners, state law societies, the accounting bodies' member directories, and state licensing registers for building, electrical and plumbing work. These are independent, indexed and nobody deletes them for being about you.
• The review platforms your customers actually use. Which ones those are is an empirical question, and the citation audit above answers it.
• Real editorial coverage. One genuine article in local or trade press does something a directory listing cannot, because it establishes a fact rather than repeating one.
• Community threads where your category gets discussed. Participate as yourself, with your affiliation stated. Astroturfing is both against every platform's rules and easy to spot.

Consistency beats volume. The same legal name, the same address format, the same phone number, everywhere. Machines resolve scattered mentions into one entity by matching facts, and two versions of your address is a reason for a system to stay quiet about you rather than risk being wrong.

What is the one source you control completely, and where does that stop?

Your own website, and it is worth about a quarter of the picture.

We publish the arithmetic rather than asserting it. In our scoring, three quarters of an AI visibility score is what the engines actually said and one quarter is site readiness. That weighting is a judgement call, and it is published on The Receipts Engine so it can be argued with.

Site readiness itself is four plain checks, worth 40, 25, 20 and 15 points: AI crawlers allowed in robots.txt, at least one JSON-LD block on the homepage, a title and meta description present, and a non-empty llms.txt. Anyone can reproduce the score by hand.

On llms.txt specifically, be careful with the hype. 71% of the 300 businesses we scanned have none, and we measured it because it says something about how legible a site is trying to be, not because it is a ranking lever. Google has said it does not use the file.

Where your control stops is worth stating plainly. You cannot edit training data. You cannot make a directory list you. You cannot delete a review you dislike, and in Australia attempting to suppress genuine negative reviews is its own problem. What you can do is make your own pages quotable, which has its own method: see how to write a page so AI can quote it.

Source: Answerable, The State of AI Visibility in Australia 2026. 300 Australian businesses, six industries, scanned 30 July 2026. Raw data published as CSV and JSON under CC BY 4.0.

See which sources AI is reading about you.

A free scan asks ChatGPT and Gemini real buyer questions in your category, then shows you what came back and which sources the answers leaned on. You keep the report.