AI Search Content That Earns Citations

Choose questions with a real audience, add original detail and make each important claim easy to check.

What a citation is and who hands them out

In Google's AI Overviews and AI Mode, a citation is a supporting link chosen from pages that are already indexed and eligible to show a snippet. Google's own guidance is blunt about it: there are no extra technical requirements, no special file and no special structured data. If a page can rank, it can be cited. Google also describes a "query fan-out" technique, where one question is broken into several related searches. That is why a page answering one narrow sub-question can appear for a broad query.

ChatGPT search works differently. OpenAI's crawler page says sites that block OAI-SearchBot in robots.txt will not appear in ChatGPT search answers, and that it can take about a day for a robots.txt change to take effect. Blocking GPTBot, the training crawler, is a separate decision and does not remove you from search.

So the first job is boring: make sure nothing is blocking the crawlers that matter, and make sure the page is indexed. Everything below assumes that.

Choose questions people actually ask

A citation follows a question, and the model can only cite you if someone asked something your page answers. For a UK small business the best sources of real questions are:

• the questions customers ask on the phone or by email before they buy;
• support tickets and the "how do I" messages that keep coming back;
• the queries already sending impressions in Search Console, including the ones with no clicks;
• questions a trade body, regulator or supplier gets wrong or answers too generally.

Pick questions where you know something specific. "What is Cyber Essentials?" is answered well by the NCSC already. "What does a Cyber Essentials Plus assessor actually sample on a laptop?" is a question a firm that has sat through the assessment can answer better than anyone.

Write so the answer can be lifted safely

Models and Google's fan-out both look for a clear, self-contained answer. Structure the page so that each important claim can be quoted on its own and still be true.

1. Put the direct answer in the first paragraph, in one or two sentences.
2. Use one claim per sentence. A sentence that mixes a fact with a caveat and an opinion is hard to quote accurately.
3. Date anything that changes. Say "at the time of writing" and name the version, plan or year. A citation of a stale price is worse than no citation.
4. Use headings that match the sub-questions a reader would ask next. Those are the fan-out queries.
5. Add detail that only comes from doing the work: the exact error message, the setting name, the form field, the time it took, the mistake you made first.
6. Name the author and say why they know. Link to the primary source for every regulatory or product claim.
7. Keep the important content in text. Google's guidance asks for text form, with images and video as support rather than as the only carrier of the answer.

Technical checks before you publish

• robots.txt allows Googlebot and, if you want ChatGPT search, OAI-SearchBot.
• No noindex on the page, and no nosnippet unless you mean it. Google's preview controls (nosnippet, max-snippet, data-nosnippet) are exactly the controls that limit what AI features can show from a page.
• Structured data, if you use it, matches the visible text. Marking up an answer that is not on the page is a policy problem.
• The page is in the sitemap, returns a 200 status, and loads without a wall of scripts before the text.
• Check the URL in Search Console's URL Inspection tool and request indexing once.

What not to do

Do not generate dozens of near-identical pages for every town or variation of a question. Google's guidance on generative AI content is that automation is fine when it serves people and not fine when the main purpose is to manipulate rankings. We cover the specific policy risk in AI content and scaled content abuse.

Do not stuff a page with quotations from other sites. A page that only aggregates is less useful to a model than the page it aggregated. Do not invent statistics or testimonials to look authoritative; a citation of an invented figure is a liability with your name on it.

Measure whether it worked

Google's Search Generative AI performance report in Search Console shows which of your pages appeared in AI features, by country and device. Referral traffic from ChatGPT shows up in your analytics with chatgpt.com as the source. Check both monthly against the list of questions you chose, and retire the ones that never appear after six months. Our Search Console measurement guide explains how to read those numbers without over-reacting.

Source basis

• Google: AI features and your website
• Google: creating helpful, reliable, people-first content
• Google: guidance about AI-generated content
• OpenAI: overview of OpenAI crawlers