AI search optimization: getting your business cited by ChatGPT, Perplexity, Gemini and AI Overviews
We make businesses citable by AI answer engines. It starts with a $1,000 audit, credited in full against the first project, then continues as a monthly retainer covering answer-first restructuring, an entity and schema graph, a 90-day refresh cadence and earned-mention work. We publish what does not work as well as what does.
What you get
A citation baseline you can argue with — We run your real buying questions through the major assistants and record who gets cited today, on what, and from which source. That baseline is the thing every later report is measured against.
An entity and schema graph — Organization, service, page and FAQ nodes that reference each other by stable identifier, so a machine can resolve that this company, this offering and this page are one entity rather than three unrelated strings.
Answer-first restructuring of the pages that matter — One question per page, answered in the first 40 to 60 words, with question-shaped headings and blocks that survive being quoted alone. Roughly 55% of AI-Overview citations come from the first 30% of a page.
A comparison content programme — Comparison content is 32.5% of all AI citations. We build the you-versus-the-alternatives surface honestly, including the cases where the alternative is the right answer, because a comparison that never concedes anything is not cited.
A 90-day refresh cadence with visible dates — A 90-day refresh cadence is associated with 4.8 times more citations, and a visible last-updated date with 1.8 times more. We run the cadence as a scheduled operation with a change log, not as an intention.
An earned-mention programme — 82% of AI citations are earned media rather than the brand's own domain, and brand mentions correlate with AI Overview appearance at about 0.664 against 0.218 for backlinks. So the work targets being written about, quoted and listed elsewhere — not just link acquisition.
AI-referrer analytics split out as its own channel — ChatGPT, Perplexity, Gemini and Copilot referrals tracked separately from organic, with conversion measured separately, because AI-referred traffic converts at roughly 4.4 times the rate of organic and averaging the two hides it.
A quarterly citation report — Which questions you are now cited on, which you lost, who displaced you and what changed. Written by a person, with the losses included.
Published pricing
AI search audit — $1,000 one-time. Citation baseline across the major assistants, entity and schema review, answer-structure review, refresh and freshness audit, prioritised fix list. Credited in full against the first project. Timeline: 1–2 weeks.
Retrofit project — Quoted from the audit. Implementing the audit's fix list: schema graph, answer-first restructuring, comparison surface, analytics split, refresh infrastructure. Timeline: 3–8 weeks.
Ongoing retainer — — then Sized to scope. 90-day refresh cadence, new answer pages, comparison content, earned-mention outreach, quarterly citation reporting. Timeline: Month to month.
Bundled with a website build — Included in the build then Care plan $300–$1,060 / month. When we build the site, the answer structure, entity graph, refresh loop and AI analytics are part of the build rather than a retrofit. Timeline: With the build.
How the work actually runs
01. Baseline (Week 1) — We write down the twenty to fifty questions your buyers actually ask, run them through the major assistants, and record what gets answered, who gets cited and which source the citation came from. We also record where you appear and where a competitor does. Output: A citation baseline document: question, current answer, current sources, your position, and the gap.
02. Diagnose (Week 1–2) — Against that baseline we audit the mechanics: entity resolution and schema, answer-block structure on the pages that should own each question, freshness and dateModified behaviour, crawlability for AI user agents, and where your brand is mentioned off your own domain. Output: The $1,000 audit: findings ranked by expected citation impact against effort, with the honest 'do not bother' list included.
03. Fix the structure (Weeks 2–8) — Entity and schema graph implemented. Priority pages restructured answer-first, one question each. Comparison surface built. Refresh infrastructure and visible last-updated dates added. Analytics reconfigured so AI referrers are their own channel from the day the work starts. Output: A restructured, instrumented site and a measurement baseline that is not retrospective guesswork.
04. Earn mentions (Ongoing) — Because 82% of AI citations are earned media, the ongoing work targets being cited elsewhere: original data worth quoting, listings and directories that assistants actually read, expert commentary, and communities where your category is discussed. Output: A running record of earned mentions and where each one has started appearing in assistant answers.
05. Refresh and report (Every 90 days) — Priority pages reviewed and updated on a 90-day cycle with the visible date and dateModified moved honestly. Then the baseline questions are re-run and the change is reported — including the questions where you lost ground. Output: A quarterly citation report: gains, losses, who displaced you, and the next quarter's plan.
Who this is for
Businesses whose category is now answered rather than searched — where a buyer asks an assistant and acts on the reply.
Companies already ranking on page one who never appear in the AI answer above it.
Teams with real expertise that is currently trapped in PDFs, sales decks and one person's head.
Who this is not for
Anyone wanting guaranteed placement in an AI answer. There is no such product, and vendors selling it are selling a guess.
Brands with nothing to say. Citation is earned by being the clearest source on a question, and clarity requires substance.
Businesses looking for immediate volume. If you need leads this month, that is paid media, not this.
The receipts behind those claims
708 pages, up from 123, on our own site — We rebuilt trainyouragent.com from 123 real pages to 708, each with a self-referencing canonical and a unique title and h1, enforced by build-time assertions. This service line sells what we ran on ourselves first.
0.664 vs 0.218 brand mentions vs backlinks — Brand mentions correlate with AI Overview appearance at roughly 0.664; backlinks at roughly 0.218. This is why the retainer buys earned mentions rather than a link-building quota.
97% of llms.txt files got zero AI-bot requests — In May 2026, 97% of llms.txt files received no requests from AI bots at all. We publish that because the honesty is the differentiator — we will add the file, we will not invoice it as a strategy.
4.3× more citations for pages over 20,000 characters — Long, dense, well-structured pages get cited materially more often than short ones. This is why our service pages are built as tables, price rows and process detail rather than a paragraph and a form.
What is answer engine optimisation and how is it different from SEO?
Answer engine optimisation is the work of becoming the source that an AI assistant builds its answer from. Classical SEO competes for a position in a list of links. Answer engine optimisation competes to be extracted, quoted and named inside a synthesised paragraph that may never show a list at all. The overlap is real: crawlability, page speed, internal linking and genuinely useful content matter for both, and any vendor telling you SEO is dead is selling something. The differences are structural. The unit of optimisation shifts from the page to the extractable block. Entity clarity — can a machine tell that this company, this service and this page are one thing — outranks keyword placement. Freshness becomes a first-class signal rather than a tiebreaker. And off-domain mentions outweigh links, at roughly 0.664 correlation against 0.218. The practical consequence is that a site can rank well and be invisible to assistants at the same time, which is the situation most businesses are actually in.
What actually increases AI citations, and what does not?
What works, in rough order of leverage: answer-first structure with one question per page and the answer in the first 40 to 60 words; comparison content, which accounts for 32.5% of all AI citations; a real refresh cadence with a visible last-updated date, worth roughly 4.8 times and 1.8 times more citations respectively; an entity-linked schema graph so a machine can resolve who you are; long, dense pages, since pages over 20,000 characters get around 4.3 times more citations; and earned mentions off your own domain, which make up 82% of citations. What does not work, as far as anyone can currently measure: llms.txt, which 97% of the time received zero AI-bot requests in May 2026; keyword density; stuffing FAQ schema onto pages that do not answer questions; and buying links, which correlate far less with AI Overview appearance than plain brand mentions do. We publish both lists because the second one is where most AEO budgets currently go, and a vendor who will not tell you what does not work is not a vendor you can measure.
How do you measure whether AI search optimisation is working?
Three instruments, none of which is a ranking report. First, a citation baseline: a fixed set of your real buying questions, run through the major assistants on a schedule, recording who is cited and from where. It is a sample rather than a census, and we describe it as one. Second, referrer analytics with AI sources split out as their own channel rather than lumped into direct or organic. This is the only first-party number in the set, and it matters because AI-referred traffic converts at roughly 4.4 times the rate of ordinary organic traffic — a difference that disappears entirely if the two are averaged together. Third, mention tracking: where your brand is being written about off your own domain, since that is where 82% of citations originate. Together these three tell you whether the work is landing. Any vendor reporting a single 'AI visibility score' with no methodology is reporting a number they made up.
Nobody can guarantee a citation in ChatGPT.
Correct, and we do not. There is no submission endpoint, no ranking dashboard and no placement to buy. What can be done is well established: make the page extractable, make the entity unambiguous, keep it fresh, publish comparisons, and be mentioned elsewhere. Then measure the citation baseline over time. Any vendor offering a guarantee is either misunderstanding the mechanism or hoping you will not check.
This is just SEO rebranded at a higher price.
The overlap is genuine and we will not pretend otherwise — good SEO hygiene is a prerequisite. The differences are structural: block-level extraction rather than page-level ranking, entity resolution rather than keyword targeting, freshness as a primary signal, and mentions outweighing links by roughly 0.664 to 0.218 in correlation with AI Overview appearance. If your current SEO vendor is already doing those four things, keep them and spend the money on media instead.
How much does AI search optimisation cost?
It starts with a $1,000 audit, credited in full against your first project. Implementation is quoted from the audit's fix list, typically three to eight weeks of work. Ongoing work runs as a monthly retainer sized to page count, category competitiveness and how much earned-mention work is in scope. There is no annual lock-in.
What is the difference between AEO, GEO and SEO?
Answer engine optimisation and generative engine optimisation are two names for the same discipline: being extracted and cited by AI assistants. SEO optimises a page for a position in a list of links. Both matter, they share technical foundations, and the practical difference is that AEO optimises the extractable block while SEO optimises the page.
How do you get a business cited by ChatGPT or Perplexity?
By being the clearest extractable source on a specific question: an answer in the first 40 to 60 words, question-shaped headings, an unambiguous entity graph, published numbers rather than 'contact us', a genuine refresh cadence, and mentions on sites other than your own — since 82% of AI citations are earned media rather than a brand's own domain.
Does llms.txt help with AI search visibility?
There is no evidence that it does. In May 2026, 97% of llms.txt files received zero requests from AI bots. It costs nothing to add and we will add one on request, but it should not appear as a paid deliverable on anybody's invoice.