AI Search Optimization · GEO / AEO
AI search optimization: get named when someone asks an AI who to hire.
A growing share of buying research now starts in ChatGPT, Perplexity, and Google AI Overviews. Those answers name a handful of businesses. This is the work of being one of them.
Generative engine optimization is the work of becoming one of the sources an AI model synthesizes when it answers a question in your category. It rests on four pillars: crawler access, extractable content structure, a resolvable business entity, and third-party citations.
The fourth pillar carries the most weight and gets the least attention. Models describe you largely through what other sources say about you, which means the highest-return work often happens off your own website entirely.
Why this matters now
The behavior change is straightforward. Someone who would previously have searched marketing agency near me, opened four tabs, and compared them now asks ChatGPT which agency they should talk to and gets three names. If you are not one of the three, you were never in the consideration set. There is no page two to be on.
This does not replace search. It sits alongside it, and the businesses that treat it as a separate discipline are getting an unusual window, because competition for these queries is currently a fraction of what it is in traditional search.
The four pillars
1. Let the crawlers in
This sounds trivial and it is the single most common failure we find. Many sites block AI crawlers, sometimes because a security plugin did it silently, sometimes because a previous developer copied a robots.txt from somewhere else. If GPTBot cannot read your site, no amount of content work matters.
The bots that need explicit allowance: GPTBot, OAI-SearchBot, ChatGPT-User, ClaudeBot, Claude-User, PerplexityBot, Google-Extended, and Applebot-Extended. There is a legitimate argument for blocking some of these if you are a publisher protecting licensed content. If you are a local business trying to get found, blocking them is self-harm.
2. Structure content so an answer can be lifted out
Models extract, they do not read appreciatively. Content that gets extracted has recognizable shape: a direct, complete answer inside the first hundred words; headings phrased as the questions people actually ask; short self-contained paragraphs that make sense without the surrounding context; explicit causal language rather than implication; and FAQ or HowTo markup that labels the structure machine-readably.
A useful test: take any paragraph out of the page and read it alone. If it still answers something, it is extractable. If it only makes sense after the three paragraphs above it, it will not be cited.
3. Build an entity a model can resolve
Models need to be confident that the business named in a directory, the one in a Reddit thread, and the one on your website are the same organization. That confidence comes from consistency: identical name, address, and phone everywhere; a connected schema graph with stable identifiers; a sameAs property listing your verified profiles; and a Google Business Profile whose details agree with the site exactly.
Ambiguity is expensive here. Two variants of your business name across the web means two half-strength entities instead of one strong one.
4. Get cited by sources the models already trust
This is where the real work is. When a model answers best marketing agency in Seattle, it is not primarily reading agency websites. It is synthesizing from directories, listicles, review platforms, Reddit threads, and local press. Those are sources with third-party credibility about you.
So the campaign includes claiming and completing profiles on the directories that already rank for your head terms, participating genuinely in the communities where your category gets discussed, earning inclusion in local roundups, and building a review base substantial enough that review platforms describe you consistently. It is slower than on-page work and it is the part that decides the outcome.
How we track it
AI visibility is measurable if you build the measurement. We define a query bank of ten to fifteen questions a real prospect would ask about your category and market, run them on a schedule through the Perplexity and OpenAI APIs, and log whether you are named, where in the answer, and which competitors appear alongside you. Over time that produces a share-of-voice trend rather than an anecdote.
Alongside it, GA4 gets segmented referral tracking for ChatGPT, Perplexity, Claude, and Gemini, so the sessions arriving from AI answers are countable rather than lumped into direct traffic.
If you want the long-form version of this method, we published it in full: how to get your business recommended by ChatGPT.
The honest caveat
AI citation behavior is unstable. Models change, grounding strategies change, and a tactic that works this quarter may not next quarter. Perplexity cites transparently and leans heavily on directories and Google Business Profile data. ChatGPT cites inconsistently depending on whether search grounding is active. Anyone presenting GEO as a settled discipline with guaranteed mechanics is overselling it. We re-validate the approach quarterly and tell you what changed.
How it runs
The engagement.
Access and structure audit
Crawler permissions, content extractability, schema graph completeness, and entity consistency across the web. Usually surfaces at least one blocking problem in the first hour.
Baseline the query bank
Ten to fifteen questions a real prospect would ask. Run them, record who gets named today. This is the number we are trying to move.
Fix access and entity
robots.txt, schema graph, sameAs, NAP consistency, Google Business Profile alignment. Fast, mechanical, and prerequisite to everything after it.
Answer-first content
Comprehensive pieces on the questions your buyers actually ask, structured for extraction, with FAQ and HowTo markup and named author attribution.
Third-party citation campaign
Directory profiles, community participation, local roundups, review base. The slow compounding work that decides whether you get recommended.
Re-run and report
The query bank re-runs on a schedule. You see share of voice over time and which competitors are gaining.
Questions
Straight answers.
How do I get my business recommended by ChatGPT?
Four things, in order of leverage. First, let the AI crawlers in: GPTBot, OAI-SearchBot, ChatGPT-User, PerplexityBot, ClaudeBot, and Google-Extended all need to be allowed in robots.txt, and a surprising number of sites block them by default. Second, structure your content so an answer is extractable: a direct response in the first hundred words, question-shaped headings, and FAQ markup. Third, build a resolvable entity: consistent name, address, and phone across the web, complete schema, and a Google Business Profile that agrees with all of it. Fourth, and most importantly, get mentioned on third-party sources these models trust: directories, Reddit, industry roundups, local press.
The fourth one does most of the work and is the one almost nobody does. Language models synthesize from what other sources say about you, not only from what you say about yourself.
Is AI search optimization different from SEO?
It overlaps heavily but diverges in two ways. Traditional SEO optimizes for a ranked list of ten links; AI search optimizes for being one of three or four sources synthesized into a single answer. And traditional SEO weights your own domain most heavily, while AI answers lean disproportionately on third-party consensus, meaning what directories, forums, and review sites say about you.
The practical consequence: you can rank fourth organically and still be the business the model names, or rank first and never be mentioned.
Can you track whether AI engines mention my business?
Yes, though it requires deliberate setup rather than a standard analytics report. We build a query bank of the questions a customer would actually ask an AI about your category and market, run them on a schedule through the Perplexity and OpenAI APIs, and record whether your business is named, in what position, and alongside which competitors.
We also segment referral traffic from ChatGPT, Perplexity, Claude, and Gemini in GA4, so you can see the sessions that actually arrive from AI answers.
Does llms.txt actually do anything?
Honestly, very little today. No major model provider has committed to reading it, and there is no evidence it influences citation. It costs about twenty minutes to write and there is no harm in keeping one current, so we ship it. But any agency selling llms.txt as a core deliverable is selling you the cheap part of the job and skipping the expensive part, which is third-party citation building.
How fast does this work?
Faster than traditional SEO in one respect and slower in another. Crawler access and content structure can change extraction within weeks, because models re-crawl frequently and search-grounded modes like Perplexity and ChatGPT search read live pages. But third-party citation building, the part that actually drives recommendation, accumulates over months.
Realistic expectation: first citations for narrow, low-competition questions within three months. Recommendation for a broad, high-value query like best marketing agency near Seattle is a longer campaign.
Want to know what it would take to rank in your market?
Book a free strategy call →