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

AI search optimization, also called generative engine optimization or GEO, is the practice of structuring your website so that AI answer engines quote it. It is not a separate discipline from SEO so much as a shift in what the finish line looks like. Classic SEO tries to win a blue link that someone clicks. GEO tries to win a sentence inside an answer that someone reads without clicking anything.

That matters because a growing share of the questions your customers used to type into Google now get answered by an AI summary at the top of the results page, by ChatGPT with browsing, or by Perplexity. In all three cases the machine reads pages, picks a handful of passages it trusts, and writes an answer around them, usually with a citation link next to the claim.

If your site never gets picked, you are invisible in that layer even when you rank respectably in the traditional one. Here is what actually moves the needle, based on how these systems select sources rather than on wishful thinking about how they should.

How do AI answer engines choose which sources to cite?

They do not read your whole site and form an opinion of your brand. The pipeline is narrower than that, and understanding it tells you what to change.

They retrieve, then they quote. An answer engine turns the user’s question into one or more search queries, pulls a set of candidate pages, and then extracts specific passages from those pages to support specific claims. The unit of selection is the passage, not the page. A great article whose useful fact is buried in the middle of a nine-sentence paragraph is harder to quote than a mediocre one that states the same fact in a clean standalone sentence.

They still depend on classic search. Most AI answers are built on top of a conventional search index. If you are not retrievable for the query, you are not in the candidate set, and nothing else you do matters. GEO does not replace the boring work of being crawlable, fast, and topically relevant. It sits on top of it.

They prefer claims that can stand alone. A sentence that carries its own subject and context can be lifted into an answer without editing. A sentence that says “this is why it fails” cannot, because the model would have to reconstruct what “this” refers to and risk getting it wrong.

They cross-check. When several independent sources agree on a fact, the model is more willing to state it. When your page is the only place on the internet making a claim, it is more likely to be summarized cautiously or skipped.

The seven changes that make a page citable

These are ordered by effort-to-payoff, and any competent person can do all seven in an afternoon per page.

1. Answer the question in the first two sentences. Directly under the heading, define the thing or answer the question in plain language, then elaborate. Answer engines heavily favor the opening of a section. If your intro is a warm-up about how the world is changing rapidly, you have given them nothing to quote.

2. Write headings as the questions people actually ask. “How much does a small business website cost?” beats “Pricing considerations.” The heading gives the retrieval system an explicit match to the query, and it tells the model what the section below is for.

3. Make your facts self-contained. Each key claim should be a complete sentence that survives being copied out of context. Instead of “it usually takes about three weeks,” write “a five-page small business website usually takes about three weeks to build.” The second one can be quoted. The first one cannot.

4. Put the specifics in. Numbers, ranges, timeframes, names of standards, concrete conditions. Vague copy is not quotable, because there is nothing in it to cite. This is also the honest test of whether you know your subject. If you cannot state a real range for what your work costs or how long it takes, that shows.

5. Add a real FAQ section. Three to six questions at the end of the page, each answered in two to four sentences that stand on their own. This is the single highest-yield structural change I make to client pages, because the format matches exactly what the answer engine is looking for: a question, then a short complete answer.

6. Mark up the page with schema. Structured data (Article, FAQPage, LocalBusiness, Organization) does not force anyone to cite you, but it removes ambiguity about what the page is, who wrote it, when it was updated, and where the business is located. For local businesses, the LocalBusiness markup and a consistent name, address, and phone number across the web do a lot of quiet work.

7. Show who wrote it and when. A named author with a real reason to know the subject, a visible publication date, and a genuine update history. Answer engines lean toward sources that look accountable. So do humans, which is convenient.

What about llms.txt, robots.txt, and blocking the AI crawlers?

Three separate questions that get mixed together constantly.

Can the AI crawlers reach you? This is the one to check first, and most sites have never looked. OpenAI, Anthropic, Perplexity, and Google all use identifiable crawler user agents, and plenty of sites block them by accident through a security plugin, a firewall rule, or an overzealous robots.txt. If you want to be cited, do not block the crawler that does the citing. Look at your robots.txt, then look at whether your CDN or WAF is quietly returning 403s to those agents.

Should you block them? Only if you have a specific reason, such as paywalled content you sell. For a small business marketing site, blocking AI crawlers costs you the visibility and buys you nothing.

Does llms.txt work? It is a proposed convention, a plain text file at your domain root that points AI systems at your most useful content. It is cheap to add and harmless. It is not a standard that the major engines have committed to honoring, so treat it as a small optional extra rather than the centerpiece of a strategy. Anyone selling you an llms.txt file as the key to AI visibility is selling you a twenty minute job.

JavaScript rendering. If your page content only appears after JavaScript executes, assume some crawlers will not see it. Server-rendered HTML is more reliably readable, which is one of several reasons I build client sites that way. That decision lives in the same family as the other website choices worth getting right.

What to stop doing

Stop stuffing keywords. Retrieval works on meaning, not on how many times you repeated the phrase. Repetition now reads as noise around your actual claims.

Stop publishing thin content at volume. Twenty shallow posts generated in an afternoon give an answer engine twenty pages of nothing specific to quote. One page with real numbers on it outperforms all of them. This is the part small businesses get wrong most often when they turn AI loose on their own blog, and it is the same pattern I described in what actually works with AI.

Stop hiding the useful part behind a form. If the substance is gated, it cannot be indexed, cannot be quoted, and cannot earn you the citation. Gate the deliverable, not the explanation.

Stop writing for the algorithm. This one sounds like a slogan, so here is the mechanical version: the systems select passages that a human editor would also consider clear, specific, and safe to quote. Optimizing for the machine and writing clearly for a person have converged more than they used to have.

How do you know whether it is working?

Honestly, measurement here is worse than in classic SEO, and anyone who tells you otherwise is overselling. Use three imperfect signals together.

Ask the engines. Once a month, put your ten most important customer questions into ChatGPT, Perplexity, and Google, and record whether you appear, who does appear, and what they said. It is manual and it takes twenty minutes. It is also the only direct observation available, and it tells you more than any dashboard.

Watch impressions against clicks in Search Console. When AI summaries answer a query, you often keep the impression and lose the click. Rising impressions with flat clicks on informational queries is the fingerprint of that shift. It is not a failure signal by itself, but it tells you where the answer is being consumed above your link.

Watch referral traffic from AI tools. Traffic from chatgpt.com, perplexity.ai, and similar sources shows up in your analytics as referrals. The volume is usually small compared to search. Watch the trend rather than the absolute number, and pay attention to what those visitors do, since people who arrive after reading a cited answer tend to arrive with the question already settled.

The realistic expectation: this is a compounding, months-long play, the same as SEO. What you are buying is presence in the layer where a growing share of buying research now happens.

Where this fits for a small business

If you have limited time, sequence it like this. Make sure the AI crawlers can reach you, since that check takes ten minutes and is occasionally the whole problem. Then rewrite the openings and headings of your five most important pages, the ones tied to how you actually make money. Then add FAQ sections to those pages. Then add schema. Then, and only then, think about publishing more.

The prerequisite under all of it is a site that is fast, server-rendered, and structured, because a site an engine cannot cleanly read cannot be cited no matter how good the writing is. If your current site is fighting you on that, my website builds run $4,500 for a standard build and $6,500 for a larger one, with a $500 per month care plan that includes the ongoing content and structure work described above. If you would rather do it yourself, everything in this post is doable without hiring anyone, which is why I wrote it out.

Frequently asked questions

What is generative engine optimization (GEO)?

Generative engine optimization is the practice of structuring content so that AI answer engines like ChatGPT, Google AI Overviews, and Perplexity quote and cite it. It overlaps heavily with SEO, because most AI answers are built from conventional search results, but it optimizes for a different unit: the individual quotable passage rather than the ranked page.

Is GEO different from SEO?

It is an extension rather than a replacement. Everything that makes a page findable in classic search still applies, because retrieval usually happens through a search index first. What GEO adds is passage-level structure: question-shaped headings, direct answers at the top of each section, self-contained factual sentences, and FAQ blocks.

How do I get my business mentioned in ChatGPT answers?

Make sure your site is crawlable by the AI crawlers, publish specific and verifiable information about what you do, structure it so individual claims can be quoted without surrounding context, and get mentioned on other sites that discuss your industry or area. Consistency of your business details across the web matters more than most owners expect, because agreement across sources raises the model’s confidence.

Should I block AI crawlers from my website?

For most small businesses, no. Blocking crawlers such as GPTBot, ClaudeBot, and PerplexityBot removes you from the pool of sources those systems can cite. Blocking makes sense mainly when your content is the product you sell.

Does llms.txt actually do anything?

It is a proposed convention rather than an adopted standard, and no major AI provider has committed to honoring it. Adding one is cheap and low-risk, so it is fine to have, but it should be the last item on your list rather than the first.

How long does it take to see results from AI search optimization?

Plan in months, not weeks. Pages need to be recrawled and reindexed before they can be retrieved as sources, and answer engines lean toward sources that have been around and are corroborated elsewhere. The structural changes are one-time work, and the visibility accrues gradually after that.

The one-sentence version

AI answer engines quote passages, not pages, so the highest-return work on your site is making each important claim a clear, specific, standalone sentence that a machine can lift without changing its meaning.

---

Jose Lugo is a CISSP-certified security engineer with 12 years of U.S. Army intelligence experience. He builds secure AI work environments and websites for businesses at josecustom.ai. See his portfolio of 13 live client systems at portfolio.josecustom.ai.