Glossary
AI visibility glossary: the terms, measurement first.
34 terms used in measuring and reporting AI visibility, each defined in one sentence, with an example and a link to the article that goes deeper.
Measurement
AI visibility
AI visibility is how often AI answer engines name your brand when buyers ask about your category, measured as a rate over many answers.
AmpliRank measures it on questions that don't mention the brand, so it shows discovery rather than recall. Engines answer each question afresh, so AI visibility is a sampled estimate and always comes with a range.
Visibility rate
The visibility rate, or named rate, is the share of answers to a defined set of questions in which an engine names your brand.
Named in 58 of 192 answers is a visibility rate of 30.2%. A rate means something only with its range and the number of answers behind it, and it is comparable only across runs that asked the same questions to the same engines.
Answer
An answer is one question asked to one AI answer engine once: the unit every visibility rate is counted in.
A run's size is questions × engines × repeats, so 25 questions asked to 3 engines once is 75 answers. Counting in answers makes runs, plans and tools comparable.
Buyer question
A buyer question is a question a potential customer would put to an AI engine while choosing a product, written the way that buyer would ask it.
A useful set covers the buying journey: the category, comparisons, pricing, security, integrations and the problems the product solves. It is written as people talk, not as a list of keywords.
Unbranded question
An unbranded question is a buyer question that doesn't name your brand, such as “best project management software for agencies”.
Unbranded questions measure discovery: whether an engine puts you on the shortlist unprompted. They are the basis of a fair headline rate. A question that names a competitor but not you is unbranded for you.
Branded question
A branded question names your brand or product, such as “is Taskmoor good for agencies?”, and shows what engines say about you once asked.
Answers to branded questions name you almost by construction, so they belong in a separate view, read for accuracy. Blended into the headline, they make the rate move whenever the mix of questions changes.
Question set
A question set is the fixed list of buyer questions a measurement asks, run after run, so that runs can be compared.
Two runs are comparable only on the questions both asked. New questions go in a new set rather than changing the old one, and each question is tagged branded or unbranded when it is written.
Run
A run is one round of asking every question in a set to each chosen engine, a set number of times.
The run is the unit of reporting: one headline rate, the competitor gaps on the same answers, and the change since the last run. Runs can be scheduled weekly or monthly.
Repeat
A repeat is the same question asked to the same engine again within a run, to capture how its answers vary.
Repeats narrow a range, but answers to one question tend to agree, so more distinct questions add more information than more repeats of a few.
Sample
A sample is the set of answers a figure is computed from; the figure estimates what far more askings of the same questions would show.
Every visibility figure is a sampled estimate, because engines answer each question afresh. A sample describes its questions, engines and dates, not the whole market.
Sample size
Sample size is the number of answers behind a figure, and it decides how wide that figure's range is.
At a 30% rate, 320 answers give a range of about 5 points either side. Halving the margin takes roughly four times the answers.
95% range
A 95% range is the interval in which a rate's true value very likely sits, given the answers collected.
Across many measurements made this way, about 19 in 20 ranges contain the true rate. Statisticians call it a 95% confidence interval. AmpliRank shows one with every rate, together with the number of answers.
Wilson interval
The Wilson score interval is the standard 95% range for a proportion, published by Edwin B. Wilson in 1927.
It stays sensible at small samples and near 0% or 100%, where the textbook “plus or minus two standard errors” gives impossible ranges. Named in 58 of 192 answers is 30.2%, with a Wilson range of 24.2% to 37.0%.
AI visibility calculator · Source: Wilson, 1927
Margin
The margin is half the width of a range: how far either side of the rate the true value may plausibly sit.
Choose the margin from the smallest change that would make you act; the number of answers you need follows from it.
Paired comparison
A paired comparison measures two rates on the same answers, so only the answers that differ between the two carry information about the difference.
It applies to two runs on the same questions and engines, and to two brands counted in the same answers. The idea goes back to McNemar's 1947 test for paired proportions.
How to compare with competitors fairly · Source: McNemar, 1947
Paired difference
The paired difference is the gap between two rates computed answer by answer, reported with its own 95% range.
In a sample of 192 answers asked in two runs, 21 answers gained a brand and 9 lost it: a change of +6.3 points, range 0.6 to 11.9.
Clear change
A clear change is one whose 95% range, paired over the same questions and engines, excludes zero; any other movement is reported as no clear change.
Overlapping ranges for two rates don't prove nothing changed, and separate ranges don't prove something did. Only the range of the difference decides.
Competitor gap
A competitor gap is the difference between your rate and a competitor's, measured on the same answers, with its own 95% range.
A gap is called clear only when its range excludes zero. Two brands 8 points apart can have a clear gap or none, depending on how the answers split between them.
Share of voice
Share of voice is your brand's mentions as a share of all mentions of the brands you track, so it depends on which brands are on the list.
With 50 mentions against a rival's 66, your share of voice is 43.1%; add a third brand named 40 times and it falls to 32.1%, with no answer changed. Report each brand's own rate next to it.
Observed vs caused
A change is observed when a measurement shows it, and caused only when other explanations are ruled out, which a visibility measurement can't do.
Engines, competitors and third-party sites change in the same weeks you publish. Reports state a change as observed since a date, with its range.
Engines and evidence
AI answer engine
An AI answer engine is a service that answers a question in written prose, often after searching the web, rather than returning a list of links.
AmpliRank supports ChatGPT, Claude, Gemini, Perplexity and Google AI Mode. The method page names the engines switched on today.
Fresh session
A fresh session is a conversation with no history or account behind it: the neutral starting point every new buyer shares.
A signed-in person's own history can move their answers, so there is no single answer to copy. Asking repeatedly in fresh sessions gives a rate with a range.
Query fan-out
Query fan-out is the set of searches an AI engine runs on its own while answering, often on sub-questions the buyer never typed.
Google says AI Overviews and AI Mode may use the technique. The searches show which sub-questions decide an answer, and so which pages are worth writing.
What is query fan-out? · Source: Google Search Central
Grounding
Grounding is an engine basing its answer on search results or documents it retrieves, rather than only on what its model learned in training.
Grounded answers can cite the pages they used. Google's Gemini documentation calls this grounding with Google Search, and its grounded responses list the searches the model ran.
Cited source
A cited source is a page an engine links to or names as a basis for its answer.
Counting which sites are cited across many answers shows which roundups, forums, publications and brand sites shape a category's answers, and so where to earn a mention.
Citation vs mention
A mention is your brand named in an answer's text; a citation is your page given as a source. Either can happen without the other.
An engine can name you while citing a review site, or cite your page while recommending a competitor. The visibility rate counts mentions; citations show which pages the engines rely on.
Reporting
Weekly brief
A weekly brief is a one-page report of a run: the headline rate with its range, the change, competitor gaps, top sources and next steps.
Each figure says whether it is a clear change or within the noise, and each next step names the evidence behind it. A free template is available to download.
Measurement cadence
Measurement cadence is how often runs happen; with a fixed number of answers a month, more frequent runs are smaller and can call only larger changes.
At 75 answers a run, the smallest clear rise from a 30% rate is about 12.0 points; at 600 answers, about 3.0, assuming 5 in 100 answers lose you by chance between runs.
AI referral visit
An AI referral visit is a visit to your site from a link in an AI assistant's answer, identified by its referrer or a tag.
Google Analytics 4 groups such visits in an AI Assistant channel, which excludes Google's AI Overviews and AI Mode. Counts are a floor: a buyer who reads an answer and returns through search is counted as search.
How do you track visits from AI answers? · Source: Google Analytics Help
Assisted conversion
An assisted conversion is a conversion with a given channel somewhere on its path, though not as the last touchpoint.
An AI-referred visit early on a path that ends in a demo request is an assisted conversion. Attribution models split credit between touchpoints by a rule; they don't measure cause.
Market terms
GEO (generative engine optimization)
Generative engine optimization (GEO) is work on content and its sources so that AI answer engines use it and name it in their answers.
The term was introduced in a 2023 research paper, “GEO: Generative Engine Optimization” by Aggarwal and colleagues, published at KDD 2024. It is now used broadly, and its scope varies by who uses it. AI visibility is the measurement that shows whether the work appears in answers.
GEO, AEO and AI visibility: what the terms mean · Source: Aggarwal et al., arXiv:2311.09735
AEO (answer engine optimization)
Answer engine optimization (AEO) is work on content so that answer engines, from search features to AI assistants, can quote it directly in their answers.
AEO and GEO overlap, and neither has a single agreed definition. Both describe work on content; AI visibility is the measurement that tells you whether that work shows up.
AI Overviews
AI Overviews are AI-generated summaries that Google Search shows for some queries, with links to the sources they draw on.
Google says they may use query fan-out, and that appearing in them needs no special optimization beyond being indexed and eligible for a snippet. Their traffic is reported in Search Console's Performance report, in the Web search type.
What is query fan-out? · Source: Google Search Central
Google AI Mode
Google AI Mode is an AI search experience in Google Search that answers with generated responses and links, aimed at exploration and complex comparisons.
Google AI Mode shows the sources it cites but not the searches it runs. Like AI Overviews, its traffic is counted in Search Console's Web search type.
How do you track visits from AI answers? · Source: Google Search Central
■ Ask · Measure · Compare · Trace · Fix · Share