In short: To be cited by AI assistants, answer one question per page in the first paragraph, make checkable claims, structure pages for machines, get the same facts stated on sites you do not own, and check the assistants monthly. The third-party step matters most and is the one most companies skip.
Key points
- A model should be able to stop after two hundred words and still have the correct answer.
- Replace words like 'leading' and 'innovative' with dated, substantiated facts.
- Render content in HTML rather than with scripts after load, so crawlers that skip JavaScript can read it.
- Keep company facts identical across trade press, partner pages, directories and review sites, because models weigh consensus.
- Blocking AI crawlers is a decision to be invisible in that assistant.
The previous article explained how AI assistants choose what to cite. This one is the list of things to do about it. None of it is exotic; most of it is the discipline good content always required, applied with a machine reader in mind.
Write pages that answer one question
Each important page should exist to answer a question a buyer actually asks, and should answer it in the first paragraph. Put the question, or a close paraphrase, in the page title and the first heading. State the answer plainly, then explain. A model extracting an answer should be able to stop reading after two hundred words and be correct.
Audit the current site for pages that do this. Most companies find that their highest-traffic pages describe the company and their most useful pages are buried three clicks down.
Make claims that can be checked
Replace "leading," "innovative," and "trusted" with facts: what the product does, for whom, at what cost, since when, with what result. Give definitions where the category is new. Give numbers where they can be substantiated, and date them. Comparisons with named criteria are among the most-cited content types, because they answer the question buyers actually have.
Structure for machines without writing for them
Clear heading hierarchy. Short sections. Lists and tables where the content is naturally a list or a table. Descriptive link text. A summary or definition block near the top of long pages. Content rendered in HTML rather than assembled by scripts after the page loads, so it is readable by crawlers that do not execute JavaScript.
Structured data markup for organizations, products, articles, FAQs, and authors helps search engines understand the page and is increasingly read by the systems that feed AI answers. It is cheap to add and rarely done well.
Get the facts stated somewhere other than your own site
This is the step most companies skip, and the one that matters most.
A model weighs consensus. A claim that appears only on your domain is one voice. The same claim in a trade publication, a partner's integration page, a comparison site, an industry directory, a review platform, a podcast transcript, a conference listing, and a press release is a pattern. Earned media, partner pages, and directories are not vanity; they are the corroboration layer.
Practical moves: ensure the company's basic facts, what it does, where it is, who runs it, are identical everywhere they appear. Pursue coverage and guest contributions in the publications the category's buyers read. Make partner and integration pages easy for partners to publish. Issue plainly written press releases for material news. Maintain accurate listings in relevant directories.
Answer the questions people ask assistants
Ask the assistants directly: what do they say about your category, your competitors, and you? Which sources do they cite? What questions do they think a buyer would ask next? Those questions, and the sources currently winning them, are the content plan.
Then write the definitive answer to each, publish it on a page built to be extracted, and make sure at least two third-party sources say something consistent.
Keep the technical basics honest
Fast pages. A clean sitemap. No important content behind logins or forms. Crawlers allowed unless there is a specific reason to block one; several AI companies publish the names of their crawlers, and blocking them is a decision to be invisible in that assistant. Canonical URLs so the same content is not competing with itself.
Measure it
Traditional analytics will show referrals from AI assistants where they pass a referrer, and branded search volume as a proxy for recall. Beyond that, the practical measure is manual: a monthly set of the twenty questions that matter most to the business, asked of the major assistants, with the answers and citations recorded. Watch whether the company appears, whether it is described accurately, and which sources are cited instead.
That is not sophisticated. It is enough to know whether the work is landing, which is more than most companies know today.
lowob takeaway: Answer one question per page in the first paragraph, make checkable claims, structure for machines, get the same facts stated on sites you do not own, and check the assistants monthly to see whether it worked.
Common questions
How do I get my company cited by ChatGPT or other AI assistants?
Answer one buyer question per page in the first paragraph, make checkable claims, structure pages clearly, and get the same facts stated on third-party sites such as trade publications, partner pages and directories.
Why do AI assistants cite other sites instead of mine?
A model weighs consensus. A claim that appears only on your own domain is one voice; the same claim across several independent sources is a pattern.
Should I block AI crawlers on my website?
Only for a specific reason. Blocking a crawler is a decision to be invisible in that assistant.
How do I measure whether AI optimisation is working?
Track AI referrals and branded search where available, and each month ask the major assistants the twenty questions that matter most, recording whether you appear, how you are described and which sources are cited.