How to get cited
by ChatGPT.

What counts as a ChatGPT citation?

A citation links an answer to a source. In ChatGPT, open the source attached to the relevant passage and inspect the page it points to. Search can produce answers with citations, but you should check the particular response you are evaluating. ChatGPT web search

For the broader terminology, see SEO vs AEO vs GEO. Here, the task is narrower: inspect one page, run one documented check and choose a useful improvement.

1. Can the search system reach the answer?

Open the published URL as a visitor first. Check that it loads, shows the useful answer and links to the correct page. Then inspect the returned HTML, indexing controls and hosting rules with your developer.

Check these four places

1. The page response. Confirm the intended URL returns the page, rather than a login, error or challenge.
2. The answer in the HTML. Check whether the important text is available in the initial response. A successful browser render alone does not establish what a crawler received.
3. Site and page controls. Inspect robots.txt, page-level robots directives and the canonical URL. Resolve exclusions or conflicting URLs deliberately.
4. The hosting layer. Check whether your CDN or firewall blocks the relevant search crawler. Use provider documentation and verified request evidence.

OAI-SearchBot is different from GPTBot

OpenAI identifies OAI-SearchBot as its search crawler and GPTBot as a crawler for content that may be used in training. Their controls are independent. Review the search rules and published IP ranges with your hosting provider. OpenAI crawler documentation

A normal HTTP request, or one that copies a bot’s user-agent, cannot prove a verified OpenAI crawler reached the page. Treat those as access diagnostics. Use verified crawler requests in your own logs when available, and keep the actual citation observation separate.

2. Make the page useful as a source

Choose a question your business can answer with substance. A setup guide can show the real setup. A pricing page can state the actual terms. An About page can identify the product and founder. Each page should make its own answer easy to understand.

Lead with the answer, then demonstrate it

Give the direct response early. Follow it with the conditions, steps, examples or comparisons the reader needs. Put evidence next to the claim it supports. Use a screenshot for a real observation, a table for a decision and a worked example for a method.

Bring verified details from your business: product behaviour, a reproducible test, an explained process or relevant first-hand material. Google’s people-first guidance asks for original value and clear authorship. These are useful editorial checks; they are not a published ChatGPT ranking formula. People-first content guidance

Help the reader identify the source

Use consistent product and company names. Link the article’s real author or editor to an informative profile. Keep structured data aligned with the visible page and connect related pages through relevant internal links.

For example, Postomatic’s homepage introduces the product promise. Its About page answers the founder question explicitly. Those pages serve different reader tasks; repeating a broad marketing headline would not answer both.

Before creating another URL, check whether an existing page already serves the question. Start with the buyer’s question to choose the task. Expand coverage when you have a distinct, useful answer to add.

3. A real check: who founded Postomatic?

On 30 October 2026, we tested a branded identity question in ChatGPT web. We supplied the domain and asked for official sources. The temporary chat was set to unpersonalized. The exact model version was not exposed in the interface.

What the response showed

The activity panel showed “Searched 6 websites”. The answer correctly named Anthony Indraus and described Postomatic’s content workflow. We opened the inline source beside the founder answer: it identified Postomatic’s About page. The response also linked to the official homepage and workflow page.

What we could inspect on that page

We separately fetched the public About page and inspected its initial HTML. This ordinary HTTP audit found:

• A 200 response and the canonical URL https://postomatic.co/about.
• The founder answer in the returned HTML, alongside a clear product description.
• Visible identity information and linked Organization and Person structured data that matched it.
• No page-level robots exclusion in that response; robots.txt allowed the site under its wildcard rule.

This establishes one successful branded citation check. It does not show unbranded recommendation frequency, a change caused by a particular edit, verified crawler access or a referred visit. The audit describes the page we could inspect; it cannot tell us why the system selected it.

4. Save an observation you can use again

Keep the same record whether the answer is favourable, missing or wrong. That gives the next check something meaningful to compare.

Question and context

The exact prompt, platform, date, available model details and personalization or search settings.

Search activity

What the interface showed. Record “not observed” when you cannot confirm search was used.

Answer and mention

Save the response or capture. Note whether the brand was named and whether the description was accurate.

Source citation

Open the source and record the exact URL, the supported claim and whether the page is yours.

Next action

Write the specific page issue, proposed change and date. Keep later observations and website visits separate.

For our example, the record is: branded founder question; ChatGPT web; 30 October 2026; search observed; correct brand and founder; inline citation to /about; no referral measured. It is one observation, so there is no visibility percentage to calculate.

Google’s AI search features are included in Search Console’s Web reporting. That does not make its clicks a ChatGPT citation count. Use what useful SEO progress looks like to keep page discovery, visits and business outcomes distinct. Google’s measurement scope

5. Choose the next edit from the evidence

Fix the issue you can demonstrate. Avoid rewriting a useful page simply because one response did not cite it.

Return to the saved prompt after the change and retain the full result. For broader discovery, test relevant unbranded buyer questions as their own set. Keep that work distinct from the founder lookup demonstrated here.