Generative Engine Optimization: How to Get Your Business Cited by ChatGPT and AI Search

Related: Migrating Off Bubble, Webflow or Wix Without Losing Your Rankings

For twenty years, search meant a list of links and the job was to be near the top of it. That is changing. A growing share of questions now get answered directly — by ChatGPT, by Claude, by Perplexity, by Google's AI Overviews — and the user never sees a list at all. If your business is not in the answer, you are not in the running.

The work of getting into those answers has picked up a name: generative engine optimization, or GEO. It overlaps with SEO but it is not the same job, and some of the habits that made sense for ranking actively hurt you here.

What changes when the answer replaces the list

A ranked list rewards relevance to a query. An answer engine is doing something different: it reads a set of sources, decides which claims it trusts, and writes a summary that cites a few of them. You are no longer competing for a position. You are competing to be the sentence the model quotes.

Three consequences follow, and they are the whole of GEO:

• Specific beats general. A model cannot quote "we offer competitive pricing." It can quote "most custom web apps fall between $3,000 and $25,000." Vague marketing copy is invisible to an answer engine because there is nothing in it to extract.
• Structure beats prose. Content that is organised into clear questions, headed sections and lists gets pulled apart cleanly. A wall of undifferentiated text does not.
• Corroboration beats assertion. Models weight claims that appear in more than one place. A fact stated only on your own site is weaker than one repeated in a directory, a review, a press mention or a forum answer.

Write facts, not adjectives

This is the single highest-leverage change, and most sites fail it. Go through your service pages and count how many sentences contain a number, a range, a timeframe, a named technology, or a specific condition. On a typical agency site the answer is close to zero.

Compare these two versions of the same claim:

Before: "We deliver projects quickly and offer excellent ongoing support."

After: "Most projects ship in three to six weeks. Support covers anything that does not work as the written scope described, at no cost, for 24 months."

The second version is quotable. It also happens to be more useful to a human reader, which is the usual pattern — GEO and good writing point the same direction far more often than SEO tricks ever did.

Be careful about one thing: only publish specifics you can stand behind. Answer engines will repeat your numbers to people who then arrive expecting them. A range you cannot honour is worse than no range at all.

Answer the question in the first two sentences

Models tend to extract from the top of a section. If a page is titled "How much does X cost" and the actual number appears in paragraph nine after a preamble about your company history, you have buried the answer.

Lead with it. Put the direct answer immediately under the heading, then explain, qualify and expand. This is the inverted pyramid that news writing has used forever, and it maps almost perfectly onto how answer engines read.

Use structured data properly

Schema.org markup gives a machine an unambiguous reading of your page rather than making it infer one. The types that matter most for a service business:

• FAQPage — the highest-value type for GEO. It hands the engine pre-formed question and answer pairs. If you have an FAQ section on a page, it should emit FAQPage JSON-LD.
• Organization — your name, logo, contact details and sameAs links to your profiles elsewhere. This is how an engine connects your site to the entity it already knows about.
• Service and Offer — what you sell and on what terms.
• Article with a real author — for anything editorial.
• BreadcrumbList — helps an engine understand where a page sits.

Two rules. First, the markup must describe what is actually visible on the page; inventing FAQs that do not appear in the content is a policy violation and gets the markup ignored. Second, validate it. Google's Rich Results Test and the Schema Markup Validator both catch the small syntax errors that silently kill an entire block.

Publish an llms.txt

The llmstxt.org convention proposes a plain-text file at /llms.txt that gives a model a curated map of your site: a heading, a summary paragraph, and linked sections. Think of it as a README for machines.

Adoption is still early and no engine guarantees it reads the file. It costs an hour to write, so the expected value is good even at modest odds. The mistake we see is treating it as a sitemap — a bare list of URLs adds nothing an engine could not crawl anyway. Put the substance in it: what you do, what it costs, how long it takes, what the terms are.

Let the crawlers in

Check your robots.txt before anything else, because this one silently undoes all the rest. The relevant agents are GPTBot and OAI-SearchBot (OpenAI), ChatGPT-User, ClaudeBot (Anthropic), PerplexityBot, and Google-Extended, which governs whether your content can feed Gemini and AI Overviews.

Note that Google-Extended is separate from ordinary Googlebot: blocking it removes you from AI Overviews while leaving classic search intact. Some publishers want exactly that. A service business almost certainly does not.

This is a genuine business decision, not a technical default. If your content is the product you sell, keeping crawlers out may be right. If your content exists to bring you clients, letting them in is the whole point.

Get mentioned somewhere that is not your own site

This is the part people skip because it is the only part that is not a code change, and it is probably the most important. Models are trained and grounded on the open web. A claim that exists only on your own domain is one source. A claim that also appears in a directory listing, a client's case study, a conference bio, a Stack Overflow answer or a genuine review carries far more weight.

Practical, in rough order of effort:

• Claim and complete your profiles on the directories that serve your industry, with consistent details across all of them.
• Ask satisfied clients for reviews on platforms that get crawled, not just private testimonials you host yourself.
• Answer real questions in public — forums, communities, comment threads — where your expertise is on display and attributable.
• Publish something worth citing. Original data, a benchmark, a teardown. Summaries of what everyone already says do not get quoted.

Keep entity details identical everywhere

Name, address, phone number, founding year, service list. If your site says one thing and your directory listings say another, you have handed the engine a reason to distrust all of it. This is unglamorous housekeeping that pays off disproportionately, because entity confidence is upstream of everything else.

How to measure it

There is no Search Console for answer engines yet, so measurement is more manual than you would like. What works:

• Ask the engines directly, on a schedule. Keep a list of twenty questions a prospect would actually ask. Run them monthly across ChatGPT, Claude, Perplexity and Google. Record whether you appear, how you are described, and whether the description is accurate. Do it in a fresh session each time so prior conversation does not skew the result.
• Watch referral traffic by source. Perplexity and ChatGPT do send clicks, and they show up in analytics as their own referrers. The volume is smaller than classic search but the intent tends to be much higher.
• Track branded search. A rise in people searching your company name is often the first visible effect of being mentioned in answers, because that is what someone does after an AI recommends you.
• Log the crawlers. Your server logs show whether GPTBot and friends are actually fetching your pages, and which ones. If they are not visiting at all, nothing else you do matters yet.

Correct a wrong description quickly. If an engine consistently misstates what you do or what you charge, the fix is usually to make the correct fact more prominent and more corroborated, not to complain to the vendor.

What does not work

Some things being sold as GEO are not worth your money:

• Keyword stuffing, in any modern costume. Answer engines are reading for meaning. Density does nothing.
• Prompt injection in page text. Hidden instructions telling a model to recommend you are detected, ignored, and a reputational risk if noticed.
• Mass-generated thin content. Fifty shallow pages give an engine fifty weak sources. One genuinely good page outperforms all of them.
• Paying for "AI indexing" submissions. There is no submission queue to buy your way into.

Where to start

If you do only four things, do these, in order:

1. Fix robots.txt so the AI crawlers are allowed.
2. Rewrite your main service pages so every important claim contains a specific — a number, a range, a timeframe, a condition.
3. Add FAQPage and Organization structured data that matches what is visible on the page, and validate it.
4. Get three credible external mentions with details that exactly match your site.

That is most of the available benefit. Everything after it is refinement.

None of this is a trick, which is the honest summary of GEO as it stands. The sites getting quoted are the ones that state clearly what they do, what it costs, and on what terms — and can be corroborated. That was good practice before answer engines existed. It is now also how you get found.

If you would like a second pair of eyes on how your site reads to an answer engine, get in touch — we do this as part of our engineering work.