How MarqIQ decides you were mentioned
Whether an answer names you is decided from the answer text by a fixed rule, never by asking an AI to grade its own answer. This article gives you that rule in full, so you can look at an answer and say why it counted or did not. If an engine gave no answer at all, read Why one AI engine returned nothing, or half your prompts were not asked instead — that is a different fault.
The two rules behind every number
- Your prompt goes to each engine word for word. There is no system framing, no instruction to be helpful about you, and no hint that your brand exists.
- The answer that comes back is read by a rule, not by a model. A model is asked to rate tone for the Sentiment tile, and nothing else.
So a mention is a fact about the text, and the same answer read twice gives the same verdict.
Which rule your name gets
MarqIQ sorts your brand name into one of three kinds and matches it accordingly. Your domain is matched alongside the name, whatever kind it is.
| Kind | What makes a name this kind | How it is matched |
|---|---|---|
| Distinctive | Anything but one plain word: a space, a digit or punctuation ("Acme Analytics", "6sense", "Yahoo!"), an internal capital ("HubSpot", "iCloud"), or no vowel ("dbt", "CRM"). | Case-insensitively, and every hit counts. No accident of prose writes these. |
| Single word | One plain word ("Rippling", "Acme", "Seismic"). | Only where it is capitalised, whatever casing you typed. A hit still has to be confirmed — see below. |
| Common word | A single-word name that is also an everyday sentence-opener ("Ramp", "Focus", "Close"), three letters or fewer, or a day or month name. | As above, and never joined by a hyphen or slash ("clay-based" is not a mention). A hit that opens a sentence must read like a product. |
Every match is bounded by letters, digits and underscores, so "Marq" never matches "marquee".
Note
Your name is matched in the casing it is written in the answer, not the casing you entered. Enter your brand as "clay", "Clay" or "CLAY" and MarqIQ looks for "Clay" either way.
What confirms a single-word name
A capital letter alone does not prove a brand. MarqIQ counts a single-word or common-word name when any one of these holds.
- Your domain appears in the same answer.
- You entered the name in capitals and it appears exactly so — "SAP".
- A company qualifier follows the hit: "Clay.com", "Ramp, Inc.", "Lever Technologies".
- The hit is a list item, a heading or a table-cell title: "1. *Clay* – …", "### Clay", "| Clay |".
- The hit is capitalised mid-sentence, where the ordinary word would be lowercase: "enrich with Clay and Apollo". This does not apply on a heading or a title-case line, where every word is capitalised anyway, nor to day and month names.
- The hit opens a sentence and reads like a name — followed by a verb, a possessive, "for", "vs", punctuation or a capital: "Rippling offers payroll".
A common-word name gets the strictest version of that last test: opening a sentence, it must also start a list of names ("Ramp and Brex are…"), or sit in a sentence that names a kind of product ("Snowflake remains the default cloud data warehouse"), or take a product verb in an answer that already names another tracked brand. "Ramp up your outbound" and "Focus is key" are none of those.
An unconfirmed hit is stored as a weak match and is not counted. On a recommendation, that engine reads — no confirmed mention, and the hover says your name appears only as an ordinary word there.
What is counted as failed rather than as no mention
A response is not always an answer. Each provider has its own way of saying the text is incomplete inside a reply that otherwise looks fine, and MarqIQ reads all of them: "answer cut off", "blocked by the provider", "declined to answer", an empty body, and "provider error mid-answer".
Take care
Those calls are counted as failed, not as "answered, did not name you". Counting a truncated answer as a miss would pull your visibility down for something the engine did, so it is kept out of the rate and reported on the run's health line instead.
Competitors, and where you rank
Every listed competitor goes through the same matcher, common-word rule included, and all the names are matched together in one pass. Your position in an answer is your rank among the tracked names, counted from where you first appear by name or by domain.
- A name is counted once per answer, however many times it appears.
- Your own brand is dropped from your competitor list, by name or by domain, so it can never compete with itself.
- Repeated spellings are folded — "Beta" and "beta" are one rival, not two.
What the excerpt shows you
Every answer is stored, and with it the passages that name you, joined with " … " and capped at 700 characters. A label line such as "2. *Acme*" takes the section beneath it, because that is where the verdict is; a long line narrows to the sentence naming you; a table row is its own passage.
The Sentiment tile reads those passages whole rather than the opening of the answer, and takes a sample of at most 32 excerpts spread across engines. A ten-item list that praises you in your entry and warns about you in a closing section is therefore rated on both.
Why a prompt naming you is not scored
"Is Acme any good?" will always mention Acme, so it measures nothing about whether an engine brings you up on its own. A prompt that names your brand, a known alias or your domain is labelled aided: asked on every run, answers shown, left out of every score.
The aided test is deliberately looser than the answer test, because a prompt is typed by a person rather than written by an engine.
- The brand as the answer matcher reads it — so a common-word name needs the same proof, and "how do I ramp up outbound" is not aided.
- Any other name or alias, whole-word and case-insensitive: "best tools like acme analytics".
- Your domain and its first label, whole-word and case-insensitive.
- Any alias of five letters or more with its spaces and hyphens moved about — "in-house marketers", "InHouse Marketers", "theinhousemarketers" — matched across runs of up to six adjacent words.
There is no alias list to maintain
The aliases are derived from your profile each time: your name without a leading "The" or a company suffix ("Acme Inc." gives "Acme"), your domain, and your domain's first label. Rename the brand or change the domain and existing prompts are relabelled on the next read.
Where the rule is knowingly imperfect
Two gaps are left open on purpose, because closing either would cost more than it saves. Both are stated here so you can spot them in your own answers.
- A common-word brand that opens its only sentence with no product reading is missed. "Ramp up your outbound" and a real "Ramp — the spend platform" are the same three letters and a capital.
- A single-word brand used as the ordinary word can be read as a mention when it opens a sentence with a verb, and a prompt using it that way is labelled aided. The aided label names the word it matched, so you can see why.
Tip
If your brand is an everyday word, put the domain in your prompts' subject matter rather than the name, and watch the — no confirmed mention chips. They are where the rule is working hardest, and where a missed mention would show first.
If it did not work
An answer talks about you at length and the run says no mention
Why: Your name is a single or common word and no hit was confirmed, or the answer was cut off and counted as failed rather than as a miss.
Fix: Read the engine's chip for that prompt on the Prompts tab. A red failed means the call gave no usable answer, so it is not counted as a miss. A grey — means the answer did not count as a mention, and a weak hit reads the same there — only a recommendation's drawer separates the two, as — no confirmed mention. Then read the answer and check whether it capitalises your name mid-sentence, names your domain, or sets it as a heading.
A mention is counted where your brand is not the subject
Why: Your name is an ordinary word that opened a sentence and read like a name — "Runway is critical for startups" reads exactly like "Stripe is a payments platform".
Fix: Read the stored excerpt on that answer. There is nothing to configure; it is the cost of matching a single-word name at all. Prompts written at category level, rather than around the word itself, produce fewer of these.
A prompt is marked aided and you did not expect it
Why: It matched your brand name, an alias derived from it, or your domain. The match is on names rather than on the literal brand string, so "In-House Marketers" and "theinhousemarketers.com" both count.
Fix: Read the reason on the row — it names the term it matched. Rewrite the question at category level and add it again.
Share of voice shows a competitor you never listed
Why: A name stored on an older answer is still counted in the breakdown for that run, even though the list has changed since.
Fix: Nothing to fix on that run. Competitors are counted from the next run, never retroactively, so run a fresh check to see the list you have now.
Next steps
Related articles
Was this article helpful?