Answer engine optimisation, explained.
Plain explanations of how AI engines decide who to recommend, and the work that earns a place in the answer. No jargon for its own sake; the precise terms, used precisely.
Vignette. A name being marked inside a generated answer. Illustration, not measurement: no figure appears in the loop.
A reference set, not a feed.
These essays explain how an answer engine picks the providers it names, and what changes about your work once it does. They are written to be read in any order, revised when the mechanism changes rather than on a schedule, and each one states when it was last reviewed.
- Definitional, not topical. Each essay answers one question and stays on it.
- No dates in the index, no author cards, no tags. None of this is news.
- The mechanism words are used precisely: retrieved, cited, indexed, answerable.
- Where a claim needs a number, the essay names the source or omits the number.
The set17 essays
- What AI share of voice actually measuresAI share of voice is your portion of the brand mentions an engine makes across a fixed set of prompts, and it is a measurement only when the prompt set, the number of runs behind it and the capture window are published alongside the number.
- Why an AI visibility score needs a measurement windowAn AI visibility figure is comparable to another figure only when both were produced by the same prompt set over a stated window, because changing the set or the window changes the number even when nothing changed in the answers.
- AI Overview presence is a rate, not a yes or noWhether Google shows an AI Overview for a query is not a fixed property of that query: it varies from one capture to the next, so presence has to be reported as a rate over repeated passes with the number of passes attached.
- Why a single composite AEO score is a defectA composite AI visibility score fuses signals whose fixes point in opposite directions, so the number cannot tell you which work to buy, and prescribing off it is how a quarter of budget goes to the wrong problem.
- How retrieval actually decides citationAn engine does not read your page and judge it: it retrieves passages from an index, fuses spans that corroborate one another, and attributes the result, which means citation is decided at the passage level before the model composes a sentence.
- What a citation rate cannot tell you about revenueA citation rate measures how often an engine names you across a stated set of prompts, and it cannot be converted into revenue, because the influence it describes usually arrives with no click for anything to attribute it to.
- Why keyword volume cannot see AEO demandKeyword volume tools report a planner corpus built from search-box queries, so a conversational question asked inside an assistant returns no data whether or not people are asking it, and a zero there describes a blind instrument rather than an absent market.
- Proving absence needs three instrumentsAn instrument that fails to find something has told you about the instrument, so any claim that a page is not indexed, an answer does not appear, or a mention does not exist needs at least three independent checks before it is a finding.
- What an AI visibility audit should containAn AI visibility audit should contain a frozen prompt set with the run count printed on every figure, a named list of the engines it reached and the engines it could not, six pillars reported separately rather than summed into a score, an explicit diagnosis of whether you are absent because you are never retrieved or retrieved and not chosen, and a fix list ordered by that diagnosis rather than by pillar.
- Why AI crawler access is the first thing to checkAI crawler access is whether an answer engine's fetcher can retrieve your pages and read your content in the raw HTML response, and it is the first thing to check because it can fail at three independent layers, robots.txt, the edge firewall and JavaScript rendering, of which only the first is visible in the file most audits read.
- Entity clarity: why AI does not know who you areEntity clarity is whether an answer engine can resolve your brand name to one specific organisation rather than to a word it already knows, and a coined name buys that clarity only after independent sources corroborate it, never on the day you register it.
- What we got wrongThis is a running, dated ledger of the measurements and changes Caldrin got wrong, kept because a control group is the only thing that separates an agency which learns from an agency with an alibi, and it opens with a page title we changed on 31 July 2026 that sent a client homepage from 16.8 to 34.5 in average position overnight while 34 untouched control pages moved 0.8.
- What an AEO agency actually doesAn AEO agency is a services firm hired to make a business more likely to be named inside AI-generated answers, and the work splits into two jobs with opposite fixes: getting your pages into the pool of sources an engine retrieves for a question, and getting selected once you are in that pool.
- How to evaluate an answer engine optimization agencyAnswer engine optimization agency, AEO agency, GEO agency and LLM SEO agency name the same service, and because the deliverable is a number rather than a ranking you can look up, the only reliable way to compare two providers is to audit the method behind their figures: the prompt set, the run count, the window, the engines covered, and what they decline to publish.
- LLM SEO agency, or the SEO agency you already haveAn LLM SEO agency is hired to get a brand named inside AI answers, and most of what it should be doing is ordinary search work aimed at a different consumer, because a page an engine cannot crawl, index or rank is a page no answer can retrieve; what is genuinely new is the structural rewriting that makes a passage quotable and the measurement layer that watches a probabilistic output instead of a ranked list.
- AI visibility agency or AI visibility tool: which one you needAn AI visibility agency is hired to measure how often AI engines name a brand and then to change it, and the choice between hiring one and buying a tracking tool turns on a single question: whether anyone on your side will act on the number, because a number is cheap to buy and acting on it is not.
- Best AEO agency: why the honest answer is a shortlist, not a nameThere is no best AEO agency, because the six providers we audited in August 2026 sell four different business models at published prices ranging from $99 a month to figures two of them decline to publish at all, and the useful output of a search like this is a shortlist of two or three providers that will take your kind of business, checked against their own published method.
Individual terms are defined in the glossary.
The inputs every essay comes back to.
One essay sets out the mechanism the rest of the set assumes: an engine names a brand it can read, trust, corroborate, and lift a clean sentence from. Those are its four sections, and each link opens at that section.
Nothing here is a measurement.
An explanation is not evidence about your category. The disclosure sits on the page rather than in a footnote, on this page as on every other.
- instrument
- None. Nothing on this page is a measurement.
- what was measured
- Nothing. These pages carry definitions and mechanism, not figures.
- how
- Written as explanations. Where a claim needs a number, the essay names its source or omits the number.
- over what window
- Each essay states the date it was last reviewed, on the essay itself.
- what this cannot tell you
- None of this tells you whether an engine names you today. That takes a prompt set run against your own category, against competitors you name, and it is measured separately.
See where you stand.
Reading the mechanism is the cheap part. The audit runs a prompt set for the questions your buyers actually ask, benchmarked against competitors you name, and the findings are explained on a call rather than emailed.