Reference
Plain-English Glossary
The terms we use, defined without jargon — and where an analogy makes something clearer than a definition does, you get the analogy too.
What we measure
The numbers on your dashboard, and what each is counting. One thing before the list, because it prevents most misreadings: mention rate counts EVERY answer, while citations, sentiment and competitive position are measured only across the answers where you appeared. Those three describe quality when you show up — not how often you show up.
Visibility Score
One number, 0–100, for how present you are in AI answers.
Built from whether and how prominently you were named, citations and sentiment, with each engine weighted by our estimate of how many people use it — ChatGPT 32%, Google AI Overview 30%, Gemini 20%, Claude 10%, Perplexity 8%. So a mention on ChatGPT counts for more than one on Perplexity. Competitive position is shown alongside the score but does not affect it. If a high mention rate sits next to a middling score, one reason can be engine mix: strong on the engines with the least weight.
Mention
An AI answer named your business.
Counted when your business name appears in the answer to a tracked phrase. The share of all answers that named you, and the number to lead with. What it does NOT tell you is position — being named eighth counts exactly the same as being the first recommendation, which is why position is reported separately, as Prominence.
Citation
An AI answer linked to your website.
A mention means the engine knew your name; a citation means it pointed at your site as the source. It matches on your DOMAIN, never your company name — for a business at example.com it looks for “example.com”, and any page on that domain counts. Naming you and sourcing from you are two entirely separate measurements, which is why the numbers can diverge sharply.
Sentiment
How positively you were described when you were named.
It reads the wording of the AI’s own response. Three things it is not: a survey of what your customers or review sites say; a measure of answers that did not mention you (absence is invisibility, never a bad review); or a large sample — on a small scan it may rest on three or four answers.
Prominence
How high up you were named in the answer.
Reported in bands rather than exact positions. The engines are non-deterministic — the same question asked twice does not return the same answer — so a precise number would imply more certainty than the measurement supports. Expect some movement between scans; a consistent direction over several scans is the signal to watch.
Priority
How much a recommendation is expected to move your visibility.
Impact, not effort and not urgency. Two problems are given the highest priority when a scan finds them, because each stops an engine using your site at all: OpenAI's search crawler being unable to reach your site, and your pages being ineligible to appear in Google's AI answers. Beyond those it is a judgment, so several recommendations can share the top band — and a high-priority item can be quicker than a low-priority one. Read it alongside the “website update required” note, which tells you who has to do the work rather than how much it matters. Note too that your list is deliberately a MIX rather than every high-priority item at once — some lower-priority actions can be quicker, and you can finish those while a slow one is still in progress. That also means a new high-priority item can appear later: the list refills as you complete things, and nothing was being held back from you.
High priority does not mean hard, and low priority does not mean pointless. One is about the size of the gain; the other is about the size of the job.
Who AI Named
The businesses the answers actually put forward, whether or not you track them.
The more trustworthy competitive read, because nobody's list biases it. Competitive position depends on which rivals you chose to track; this reports whoever genuinely appeared.
Competitive position
How often you appear relative to the competitors you track.
100% means you appeared every time a competitor did. Treat it carefully: it depends entirely on which competitors you track, and with few or none tracked it says very little. It is also not offered as something to act on directly — it moves as a consequence of the other measures shifting relative to everyone else.
Read mentions and citations together
This is the most informative thing on your dashboard, and neither number says it alone.
AI knows you and is reading your site. The healthy case.
Engines name you but do not link to your site. They may be reading other pages about you, or answering from what they learned in training; from the outside the two look the same.
You are being read but not recommended.
The middle case is the one worth stopping on: your own site is not the page engines are using when they describe you.
How AI engines find you
The part that causes the most confusion, because two completely different systems both end up feeding the same answer.
Search phrase
A question someone asks before they know your business exists.
Not your business name, and not how you describe yourself. The phrases you track decide what your score actually measures.
A phrase whose answer cannot name any business cannot tell you apart from your competitors. “How much does a home theater cost” names nobody, so it scores zero for everyone.
Crawler (bot)
A program that visits web pages automatically.
Different crawlers do different jobs, and they announce themselves by name so a site owner can allow one and refuse another. OpenAI runs three: OAI-SearchBot for search, GPTBot for training, and ChatGPT-User for live fetches when someone asks a question. Anthropic runs the same three-way split for Claude.
OpenAI — crawlers and site owner controls · checked 29 August 2026
Anthropic — web crawling and site owner controls · checked 29 August 2026
Index
The stored copy of the web an engine searches when you ask it something.
Built in advance by crawlers, on their own schedule. When you ask a question, the engine searches this stored copy; some engines can also fetch a page at the moment someone asks, like ChatGPT-User above.
Retrieval
The engine looks something up and reads it back to you.
The loop you can influence. Publish a change and it can appear once a crawler has read the page again; no vendor publishes when that will be. indextr scans with live search switched on.
A librarian with a photocopier. It walks the web copying pages into a filing cabinet. When you ask a question it finds the right page, pulls it out and reads it aloud — which is why you get a citation, and why fixing your page changes the answer next time it is copied.
Training data (parametric knowledge)
What the model absorbed when it was built, and can recall with no search at all.
The slow loop. It changes only when a new model is trained, and nothing you publish this week affects it.
A student reading for a degree. It reads enormous amounts, then answers from memory — but cannot tell you which book a fact came from, which is why there is no citation. Handing it a new book today changes nothing, because it already graduated.
Recrawl
A crawler coming back to see what changed.
The gap between making a change and an engine noticing it. No vendor publishes a schedule, which is why time-to-impact estimates are estimates.
Fixing a broken doorbell tells you the door works. It does not tell you the postman has been.
robots.txt
A file telling crawlers which parts of your site they may visit.
It is a request, honoured by well-behaved crawlers. Worth knowing that blocking is not always done here: a firewall, security service or CDN can block a crawler by name before a request reaches your site, without the site owner noticing. indextr checks OpenAI's search crawler on every scan.
Things on your website
What an AI answer can quote, and what it cannot.
Title tag
The page's name, shown in the browser tab and in search results.
A title reading “Home” says nothing about the business. It should carry the business name and what distinguishes it — what you do, and where. A note on terminology: people often say “meta tags” for the title and description together, and any developer will know what is meant, but strictly the title is its own HTML element and not a meta tag at all. “The title tag and meta description” is the precise phrasing.
Meta description
A one-sentence summary of the page, in the HTML.
Plain text an engine can read and quote, and a genuine meta tag. If you do not write one, engines assemble a summary from whatever text they find, which produces exactly the stitched-together fragments you would expect. Keep it short: search results cut long ones off.
Alt text
A written description of an image, for people who cannot see it.
Also the part of an image an engine can read as text. Google states it uses alt text alongside the page's content to understand what an image shows.
Google Search Central — Google Images best practices · checked 28 August 2026
Schema markup (structured data)
Machine-readable code declaring facts about a page — and three words people use interchangeably that are not the same thing.
The three terms nest. STRUCTURED DATA is the umbrella — any machine-readable markup describing what a page is about. SCHEMA.ORG is a vocabulary, the shared dictionary of types like LocalBusiness, FAQPage and Review; it is so dominant that people say “schema” when they mean structured data generally. JSON-LD, MICRODATA and RDFa are formats — three ways of writing that vocabulary into a page. Google says it “recommends using JSON-LD for structured data if your site’s setup allows it”, which is a preference among formats for sites that choose to add it, not a statement that you need any.
indextr does not recommend adding structured data for AI visibility. We checked all five engines in August 2026 and not one documents a requirement for it; Google states outright that there is “no special schema.org structured data that you need to add”. We also do not recommend removing what you already have, because it does real documented work for Google rich results — the enhanced listings in ordinary search. A lower priority than it once was, not a mistake to be undone.
Not to be confused with the title tag and meta description, which we DO recommend writing. Those are page text an answer can quote. Structured data is a separate claim attached to the page.
Google Search Central — AI features and your website · checked 29 August 2026
Schema is a claim ABOUT the page. The title and the words on it ARE the page. An answer can quote the second.
Citation source
A page on someone else's site that mentions your business.
Which sources matter depends on the engine. Across 3,490 citations in our scans to 5 September 2026, every engine drew most of its citations from ordinary websites, mostly businesses' own (73% to 86% depending on the engine). Google AI Overview leaned on Business Profile cards (12%); YouTube pages appeared on Gemini (7%) and Google AI (5%); Yelp appeared mainly on Claude (6%).
Credential (certification or membership)
A qualification you hold, usually displayed as a badge image.
Shown only as an image, it is not text an answer can quote. Put the credential in visible text, give the badge real alt text, and link it to your profile on the issuing body's site.
A photograph carries no text. The badge is a picture of a fact, and a picture is not something an answer can quote.
See these numbers on your business
Run a free scan and find out whether ChatGPT names you. Paid plans add up to four more engines.
Get started free