What AI search actually does
AI assistants do not rank pages the way a classic search results page does. They retrieve information from an index or the live web, interpret the question, synthesise an answer and sometimes cite sources. Different assistants use different indexes, browsing tools and citation rules. Treat "AI search" as a set of discovery surfaces, not one ranking system.
The practical goal is to be understandable, retrievable and citable. That is closer to good technical SEO plus clear writing than to a secret prompt trick. The same page can be ignored by one assistant and quoted by another, so measure across surfaces rather than chasing a single result.
Make your entity unambiguous
An AI system needs to know who you are, what you sell, who it is for and how your name is written. Use one product name, one short description and one canonical URL everywhere. Keep the company name, logo, founder information and social profiles consistent across your site, launch boards, directories and review sites.
Add Organization and WebSite structured data on the homepage. On product pages, describe the product in plain text and use Product or SoftwareApplication markup only when the details are genuine. Inconsistent names and descriptions create ambiguity; ambiguity makes citations less likely.
Write the homepage as if a new reader has never heard of you: what it is, who it is for, what it replaces and how to start. Do not make an assistant infer the category from a clever tagline.
Make pages retrievable
Many assistants rely on crawlers and indexes. If a page is blocked, client-rendered, behind a login or missing from internal links, it is a poor citation candidate. Check:
robots.txtdoes not block the crawlers you want to allow.- Important copy is in the server-rendered HTML.
- Pages load fast and do not depend on JavaScript to show the main content.
- Canonical URLs are correct and sitemaps are current.
- Key pages are linked from the homepage, hub pages and other articles.
AI crawler user agents and policies change. Review the official documentation for each provider and the Google Search Central guidance rather than copying a list from an old blog post. If you block a crawler, you remove the possibility of that assistant citing you. Decide deliberately.
Write answer-shaped content
Assistants quote passages that answer a question cleanly. Lead with the answer, then add context. A useful pattern:
- H2 as the question. Use the wording a buyer would use.
- A direct answer in one or two sentences. Do not warm up for three paragraphs.
- A short explanation, example, table or steps. Make the answer usable.
- A source or a note about the method. Explain where a claim came from.
- A date if the answer changes over time. Pricing, integrations and regulations go stale.
Use definitions, comparisons and checklists. Avoid vague claims that cannot be verified, and avoid repeating the same paragraph under five different headings.
Prove what you say
Citations go to pages that are specific and trustworthy. Show your methodology, link to primary sources, date the page and explain who wrote it. If you publish a comparison, state the criteria. If you publish a number, say where it came from.
Do not invent testimonials, customer counts, rankings or results. AI systems and readers both have reasons to discount them, and inflated claims can outlive the campaign that produced them.
A visible editorial policy and an author or team page are small signals, but they help a reader decide whether the page is worth citing. Consistency between the page, the structured data and third-party profiles matters more than a long list of trust badges.
Use structured data carefully
Structured data helps machines parse a page, but it does not guarantee a rich result or a citation. Use the types that describe what is actually on the page: Organization, WebSite, BreadcrumbList, Article or BlogPosting, FAQPage, Product and SoftwareApplication. The definitions and expected properties live at schema.org.
Keep the markup consistent with visible text. Do not mark up a FAQ that a visitor cannot see, and do not claim review ratings that do not exist. Invalid or inflated markup is worse than no markup.
JSON-LD is the least fragile format for most teams. Validate it with a schema validator after changes, and keep the generator in code so every page uses the same rules.
llms.txt and other machine-readable indexes
llms.txt is an emerging convention: a plain-text summary that points AI tools to canonical pages, indexes and feeds. It is not an official standard supported by every provider. Publish one if it is cheap to maintain, keep it accurate, and do not treat it as a substitute for crawlable HTML.
The same applies to JSON indexes, RSS feeds and API catalogues. They are useful discovery aids. The underlying pages still need to be clear, original and maintained.
Our own machine-readable resources are listed at llms.txt. They are generated from published content, not hand-maintained copies that drift.
Measure AI visibility
AI attribution is imperfect. Use several signals:
- Referral traffic from assistant domains in analytics.
- Branded search and direct traffic around launches.
- Server logs showing AI crawler activity.
- Manual prompt checks for your category, run consistently.
- Mentions and citations found by searching for your product name plus a category question.
Track changes over time instead of expecting a single dashboard. Record the prompts, dates and answers so you can compare. If an assistant cites a competitor and not you, look at what that page does better: clearer definitions, stronger sources, a more complete comparison or a more accessible page.
Frequently asked questions
Is AEO different from SEO?
They overlap. SEO is the broader discipline of being found in search. Answer engine optimisation (AEO) focuses on directly answering questions in a format that search engines and assistants can quote.
What is GEO?
Generative engine optimisation (GEO) is the practice of improving the chance that a generative AI answer includes or cites your brand. It relies on retrievable pages, clear entities, useful original content and consistent third-party mentions.
Does llms.txt guarantee AI citations?
No. llms.txt is an emerging convention, not a standard supported by every AI provider. It can help discovery, but crawlable HTML, clear writing and a maintained site matter more.
Should I block AI crawlers?
If you want to be cited by an assistant, blocking its crawler removes that possibility. Review each provider’s documentation and decide based on your content and business model.
How do I know whether AI cites me?
Use analytics referrals, server logs, manual prompt checks and searches for your brand plus category questions. No single tool captures every AI citation, so combine signals over time.
Can I optimise a product page for AI search?
Yes. State what the product is, who it is for, what it costs and how it works. Add a short FAQ, keep entity details consistent and use structured data that matches the visible page.