Cited Layer › Services › AI search optimization
Service
AI search optimization
AI search optimization is the work of getting a brand named inside the answers that ChatGPT, Perplexity and Google AI Overviews give to buying questions — rather than ranked on a page of links.
I do it as a six to eight week engagement for a single business at a time. It starts with a measured baseline, ends with the same measurement repeated in the identical mode, and the report states what moved and what did not.
Independent consultant, not an agency. Everything is delivered in writing; no call is required.
The discipline is also called GEO, generative engine optimization, or AEO, answer engine optimization. The labels are interchangeable. The methods behind them are not.
The mechanism
Two kinds of question, two different jobs
Across twenty tracked purchase queries on a live store, one split held without a single exception, and it decides what work is worth doing.
Questions of fact
Who delivers to this city today? What is in stock? What does it cost? These are answered from your own pages. Publish the checkable fact and you get named. This is winnable directly, and quickly.
Questions of judgment
Which brand is best? Who is trustworthy? These are answered from other people's pages. Your own site barely participates. Winning here means getting into the sources the engines already quote — a slower, different job.
Most of the money spent on AI visibility is spent writing content for judgment questions on the brand's own site, where it cannot work. The first week of any engagement is spent establishing which of your questions fall on which side, because that determines everything after it.
Where the answers come from
The citation layer is category-specific, and it must be measured, not assumed
Two categories measured three days apart, same protocol, opposite results:
Consumer electronics, UAE
Around twenty third-party domains carry the category: local ranking guides, price aggregators, one review platform. A poor review score on a third-party site changed the wording of an answer directly, downgrading a major retailer inside the text.
GEO services, English
About ninety per cent of citations are the vendors' own websites. Across sixty-odd answers there was not one review platform and not one forum thread. Building reviews for citation here would be wasted work.
The same prescription applied to both categories would fail in one of them. This is the single most common reason a GEO engagement produces nothing: a generic playbook, applied without measuring where that particular category's answers are sourced from.
Engagement
Six to eight weeks
-
Week 0 — baseline and prediction
Your questions, three runs each across two engines, one fixed mode, all answers recorded. Server logs analysed for which AI crawlers actually reach you and which URLs they take. A written prediction of what will move and what will not, registered before any work starts.
-
Weeks 1–2 — the homepage
On a measured site, the agent that fires when someone asks a live question took the homepage on 84% of its visits. That makes the homepage the answer surface, not the brochure. Assortment, prices, delivery windows, stock, guarantees, brand names — in plain checkable text.
-
Weeks 2–3 — category and product pages
Facts per page: price, availability, delivery time. Engines read stock status and warn the buyer inside the answer, so a stale out-of-stock flag actively costs you sales.
-
Week 3 — the technical layer
Explicit crawler permissions, structured data, sitemap, IndexNow. One day of work, and it is necessary — but on its own it moves nothing, and I will say so rather than bill it as a strategy.
-
Weeks 4–6 — the sources the engines quote
Only if the baseline showed a third-party layer in your category. The target list comes out of your own measurement: the domains actually cited in your answers, ranked by frequency, which is usually a short and finite list.
-
Weeks 6–8 — re-measurement
The same questions, the same runs, the same mode. The report includes the failures: on the reference deployment, thirty pages built for one content pattern earned zero citations in eleven days, and that is written up as plainly as the wins.
Evidence
The reference deployment, and my own zero
An eight-week deployment on a bootstrapped specialty e-commerce store in the UAE, no ad budget: ten of twenty tracked purchase queries reached the top five in ChatGPT from a baseline of zero, including first place on the trust query for the category, with all four control queries behaving as predicted. The site's own phrasing came back verbatim inside answers.
The same page should tell you the other half. This site was measured against ten of its own buyer questions and scored zero out of ten in both engines. The prediction that it would score zero was written down before the measurement was run, and the re-test date is fixed. Six of those ten questions return no vendor list to anybody, which is exactly the finding the method is built to produce early.
A consultant whose own site is not yet cited is either honest about it or hiding it. The measurement is published here for the same reason I will hand you your raw answers: numbers you cannot check are not evidence.
Limits
Three things I will not sell you
A ranking, or a "top 3 in ChatGPT"
Positions do not exist in answer engines. The same question, asked in the same minute on two accounts in an identical mode, returned a five-name shortlist to one and no shortlist at all to the other. What is measurable, and what I report, is the share of answers that name you on a fixed question set.
An llms.txt strategy
Across roughly 2,100 verified crawler fetches on two sites, the number of requests to /llms.txt from GPTBot, ChatGPT-User, OAI-SearchBot, PerplexityBot and ClaudeBot was zero. Only bingbot and YandexBot ever fetched the file. Publishing one is harmless; charging for it as a mechanism is not honest, and I removed that claim from this site when the logs contradicted it.
Pages built to match the shape of a query
Thirty pages of the "where to buy X in [city]" pattern earned zero citations in eleven days on a site that was being cited elsewhere in the same period. It is a dead layer for purchase questions. Guides that answer how to choose between products do get cited — the distinction is measurable and it is not the one most playbooks make.
Price
Four ways in
| What | Scope | Price |
|---|---|---|
| Category check | Four buyer questions, two engines, eight runs. Tells you whether your category returns vendor lists at all. Two working days. | Free |
| Visibility audit | Ten questions, sixty recorded answers, source and crawler analysis, written report. | $400 |
| Audit & first sprint | Audit, two weeks of implementation, re-measurement in the identical mode. | $1,800 |
| Full deployment | Six to eight weeks, baseline to re-measurement, report including what did not move. | $6,500 |
| Retainer | Monthly measurement and implementation, after a deployment. | $2,000–2,800/mo |
Questions
Frequently asked
What is AI search optimization?
The work of getting a brand named and cited inside the answers AI engines generate for buying questions, rather than ranked in a list of links. In practice it means publishing checkable facts where the engines actually retrieve them, making the site machine-readable, and — where the category requires it — appearing in the third-party sources the engines quote.
Is AI search optimization different from SEO?
The technical foundations overlap almost completely, and anyone claiming otherwise is selling novelty. What differs is the target and the measurement: a citation inside an answer instead of a position on a page, measured as a share of repeated runs instead of a rank.
Which engines are worth optimising for?
It depends on what you sell. For a product business ChatGPT carried the great majority of measured arrivals. For a service business it is the opposite: ChatGPT returns a vendor list on only a couple of phrasings, while Perplexity names providers freely on nearly all of them. That is measured, not assumed, and it is the first thing the baseline settles.
How long before anything changes?
Weeks, not days. Between publishing a change and seeing it in answers sits crawl, index and retrieval. On the reference deployment a measurement taken eleven days after a large publication showed nothing from that layer, while other work from the same period was already being cited. Every engagement includes a free follow-up measurement four weeks after the final one for this reason.
Do you work with agencies?
Yes, on a white-label basis. The deliverable is a written work order your developers execute and I verify from outside — no access to your systems is required for the technical layer.
What if nothing moves?
The engagement defines that case before it starts. If the questions qualified as winnable show no increase in presence rate, the next sprint is unbilled. Questions I flagged as unwinnable at baseline are excluded from that guarantee, which is why they are named in writing before you pay.
Find out whether your category is winnable
Send your site and three or four questions your buyers actually ask. The free check comes back in two working days with the eight runs, the names that appeared instead of yours, and the domains the engines cited.
alex@citedlayer.com