People increasingly ask an assistant a question and read one synthesised answer instead of ten blue links. If your business is not part of that answer, you are invisible at the moment of decision. Generative engine optimisation, sometimes called answer engine optimisation, is the work of making your site easy for these systems to retrieve, understand and cite. This guide covers the practical steps, in the order we apply them.

Traditional search ranks pages for a query. AI assistants retrieve passages from several sources, combine them and often cite a few. That shifts what matters: a page needs to contain a clear, self-contained answer, come from a source the system trusts, and be fetchable without running complex scripts. It also means that a mention without a link can still influence a buyer, so brand consistency across the web matters.

1. Write answer-first content

Start each important page with a direct 40 to 60 word answer to the question the page addresses. Put the definition, the verdict or the number first, then the detail. Assistants lift the passage that answers the question; if the answer is buried under a long introduction, a competitor's page gets the citation.

  • Use question-style headings that match how people ask: "How much does custom ERP development cost?"
  • Follow with short paragraphs, lists, tables and step sequences.
  • State assumptions and ranges honestly instead of vague marketing language.
  • Add a concise FAQ at the end of each page with real questions.

2. Make everything server-rendered and crawlable

Many AI crawlers do not execute JavaScript, or do so poorly. Make sure the content that matters is present in the HTML delivered by the server. Check by fetching the page with a plain HTTP client and reading the response. Keep navigation as real links, avoid hiding content behind interactions, and provide text alternatives for diagrams.

3. Add structured data that matches the page

Structured data tells machines what a page is about. Use Organization and WebSite markup sitewide, Service or SoftwareApplication on product pages, Article on posts and FAQPage where a visible FAQ exists. Connect entities with stable identifiers so the organisation referenced on one page is the same entity on every page. Markup must reflect visible content; do not mark up claims that are not on the page.

Avoid adding self-serving review ratings to your own organisation markup. They are not eligible for rich results and risk a manual action.

4. Publish llms.txt and Markdown versions

The llms.txt convention is a Markdown file at the root of your site that summarises who you are and lists your most important pages with one-line descriptions. Pair it with clean Markdown versions of key pages, available by adding .md to the URL. They are cheap to generate from your content store and give tools a noise-free copy to read. Support among providers is uneven, so treat it as inexpensive insurance rather than a guarantee.

5. Set a deliberate crawler policy

Different bots do different things: some crawl for model training, some retrieve pages live to answer a question, and some act on behalf of a user. Decide per bot whether you want access. In robots.txt you can allow or disallow named agents such as GPTBot, ClaudeBot, PerplexityBot and Google-Extended individually. Document the policy in a human-readable page and review it periodically, since bot names and behaviours change.

6. Keep your entity information consistent

Assistants build a picture of your company from many sources. Use the same name, description, address and social profile links everywhere: your site, directories, review platforms and press mentions. Make a short "facts" block, what you do, who you serve, where you operate, how long you have been doing it, and reuse it. Inconsistency makes it harder for a model to be confident, and hesitant systems cite someone else.

7. Build authority the old-fashioned way

Citations favour credible sources. Publish original experience: case studies with specifics, honest comparisons, and guidance written by named practitioners. Show author names, review dates and sources. Earn mentions from reputable sites in your field. None of this is new, and it remains the largest factor in whether you are trusted enough to be quoted.

8. Measure what you can

Attribution from AI assistants is imperfect, but you can see part of it. Track referrals from known assistant domains such as chatgpt.com, perplexity.ai, claude.ai, gemini.google.com and copilot.microsoft.com as their own traffic source. Add a custom dimension for it in your analytics. Periodically ask the assistants the questions your customers ask and record whether you are mentioned, how you are described and which pages are cited.

  1. List twenty questions a buyer asks before contacting you.
  2. Check each across several assistants every month.
  3. Record mention, accuracy and cited URL.
  4. Fix the page or the entity data when answers are wrong.

Content patterns that get quoted

Assistants quote passages that are complete without context. Write paragraphs that can stand alone: name the subject in the sentence instead of saying "it", include the qualifier that makes the statement true, and keep the claim and its caveat together. A paragraph that begins "For a standard ERP rollout with up to ten integrations, discovery usually takes two to four weeks" is far more quotable than one that says "this typically takes a few weeks".

  • Definitions: "X is..." sentences near the top of a page.
  • Comparisons: tables that compare options on named criteria.
  • Procedures: numbered steps with a clear outcome for each.
  • Numbers with assumptions: ranges that state what they assume.

Common technical mistakes

Several avoidable problems keep sites out of AI answers. Content rendered only by client-side JavaScript. Blocking all unknown bots in a firewall rule that was meant for scrapers. Duplicate pages without canonical tags, which split signals. Missing or contradictory structured data. Pages that return a successful status code for errors. A short technical audit usually finds two or three of these and fixing them is cheaper than any content campaign.

A simple monthly routine

  1. Review server logs for AI crawler activity and errors.
  2. Check that llms.txt and sitemaps reflect current pages.
  3. Run your question list across assistants and note changes.
  4. Update the pages with outdated or incorrect answers and republish with a visible update date.

A realistic expectation

GEO is not a trick. It is clear writing, technical hygiene and consistent facts, applied with the machine reader in mind. The work compounds, and much of it improves classic search results as well. This website was built with these practices from the first page: answer-first intros, a connected schema graph, llms.txt, Markdown mirrors and an explicit bot policy.

If you want help applying this to your own site, see our SEO and GEO services, or ask for a technical review through our contact page.