Competitor tracking

See who else is in the answer, on your own denominator

Add a competitor by domain and Attensira re-reads the answers it already collects, looking for that brand. No extra runs are issued, no prompt slots are consumed, and the competitor's mention and citation rates are computed over exactly the same successful runs as your own — which is why they do not sum to 100%.

Get Started Book A Demo

By Karl-Gustav Kallasmaa, founder of Attensira · Last updated 3 September 2026

In short

Competitor tracking scores rival brands over the answers your prompts already produced. You get the same two rates you get for yourself — mention rate and citation rate — on the same denominator, plus each domain's share of all citations in the window. Adding a competitor issues no new runs and costs no prompt slots.

Added by domain

The domain makes a citation attributable; the brand name makes a mention detectable. Attensira uses both.

No extra runs

Competitors re-read answers you already have, so they consume no prompt slots and issue no new readings.

Rates that overlap

One answer can name four brands. Every rate uses your denominator, so they routinely total well over 100%.

No invented ranking

No sentiment, no brand position, no share of a fixed pie. The product refuses to produce numbers it cannot observe.

The problem it solves

An assistant answering "what are the best expense tools for a small agency?" does not pick a winner. It lists five. Every competitive tool you have used implies the opposite — share of voice, market share, ranking position, all dividing a fixed pie that sums to one — and none of those shapes fits a paragraph that names four products at once.

So the useful competitive question is not "what slice do I own" but "how often is each brand in the room, and when it is, whose page did the model link?" Competitor tracking answers exactly that, using the answers you are already collecting, and refuses to answer the question it cannot observe.

How it works

You add a competitor by domain — example.com — rather than by display name. Adding one tells Attensira to look for that brand in the answers it is already collecting. Every answer from every tracked prompt is re-read for the competitor's name and domain, and the result becomes a competitor mention rate and citation rate next to your own.

Mention rate — successful runs that name the competitor, divided by all successful runs in the window.
Citation rate — successful runs that attach a URL on the competitor's domain, divided by the same denominator.
Source share — what fraction of all citations in the window went to that domain. Unlike the rates, source shares do describe parts of a whole.
Average citation position — the mean index of a URL within a model's citation list, computed per source domain. Not a rank.

Because the analysis runs over answers you already have, a competitor added today is measured from that point forward, on the same daily cadence as everything else. There is no backfill. Competitors do not consume prompt slots, and you can add one programmatically over the MCP server with the add_competitor tool — a read-and-write action that spends no credits.

A worked example

Ten successful runs of one prompt. Here is what the answers contained:

Run  1: names You, Acme, Borealis
Run  2: names Acme
Run  3: names You, Acme
Run  4: names Borealis
Run  5: names You, Acme, Borealis
Run  6: names Acme, Borealis
Run  7: names nobody tracked
Run  8: names You
Run  9: names Acme, Borealis
Run 10: names You, Acme

Brand	Runs naming it	Mention rate	Denominator
You	5	50%	10 successful runs
Acme	7	70%	10 successful runs
Borealis	5	50%	10 successful runs

That totals 170%, and nothing is wrong. Acme appears in seven of ten answers, you appear in five, and you appear together five times. Run 7 named nobody tracked, and it still sits in every denominator — an answer that ignored the whole category is real evidence, not a missing data point.

The practical consequence is worth internalising: your mention rate going up does not require a competitor's to go down. You can both rise, if models start naming more products per answer. Read each rate as how often this brand is in the room, never as what slice of the room it owns.

Compare like with like. Both rates come from the same runs, the same prompts, the same surfaces and the same days, so a difference between two brands on one topic is a real difference. A difference between two topics is not, because the prompts differ. And watch the counts: a competitor at 100% off three runs and one at 62% off eighty are not comparable claims.

What it does not do

No sentiment. Nothing scores whether an answer spoke about a brand kindly. If you need to know that, read the stored answer.
No brand rank or position. Neither exists in the engine, and the number that looks like one — average citation position — is a per-domain index inside a citation list.
No share of voice that sums to one. Splitting an answer that names four brands would invent information the model never gave. A wire field called ShareOfVoice holds the brand's own mention rate under an older name.
No backfill. A competitor added today is measured against answers collected from today forward.
Detection cuts both ways. Mentions are found by case-insensitive substring match with no word-boundary check, so a competitor whose name is a common word carries systematic noise — and the comparison is unfair in their favour. Read a few stored answers before trusting a rate.
It is not traffic, revenue or a ranking. It is the frequency with which a model, asked your buyers' questions, brings a brand up.

Changes are gated the same way everywhere in the product. A competitor's movement is only reported once it clears a two-proportion z-test at 95% confidence; below that you get no proven change, with a null value rather than a zero. Do not read that as flat, and do not put it in a board deck as zero.

Which plans include it

Competitor tracking is on every plan, and competitors never consume prompt slots or issue extra runs. What your plan governs is the evidence underneath: how many prompts you track, how deeply each is sampled, which model surfaces answer them, and how long the readings are kept.

Plan	Per month	Prompt slots	Sampling depth	Model surfaces	Retention
Starter	$129	50	n=1	ChatGPT, Perplexity, Claude, Google AI	3 months
Growth	$349	150	n=3	ChatGPT, Perplexity, Claude, Gemini, Google AI	12 months
Business	$699	350	n=3	ChatGPT, Perplexity, Claude, Gemini, Google AI	Unlimited

Yearly billing is charged at ten months' price. Full plan detail is on pricing.

Frequently asked questions

Does adding a competitor use up my prompt slots?

No. Competitors are scored over the answers your existing prompts already produce. No extra runs are issued, no prompts are consumed, and adding a competitor does not change your sampling depth or cadence.

Why don't competitor rates add up to 100%?

Because a single answer can name several brands at once, and every rate is measured against the same denominator — all successful runs in the window. Each brand is scored independently, so the rates overlap by design and routinely total well over 100%. They can also total under 100%, when many answers name nobody you track.

Can I see a competitor's history before I added them?

No. The analysis runs over answers Attensira collected, so a competitor added today is measured from that point forward, on the same daily cadence as everything else. There is no backfill, exactly as with a new prompt or a newly added model surface.

Do you measure sentiment about competitors?

No. Attensira does not measure sentiment for anyone, including you, and it produces no brand position or rank. Neither number exists in the engine. The verbatim answer for every successful run is stored, so if you need to know how a brand was described, you can read what the model actually said.

How should I compare myself to a competitor fairly?

Compare within the same topic and window, and read n before the percentage. Both rates come from the same runs, prompts, surfaces and days, so a difference on one topic is real; a difference between two topics is not, because the prompts differ. And check whether the change cleared the 95% significance gate before calling it a trend.

Can I add competitors programmatically?

Yes. The MCP server exposes an add_competitor tool, and the REST API is its deliberate twin. The credential names the workspace, so a request cannot be retargeted at somebody else's data, and adding a competitor is not one of the actions that spends credits.

You set the destination. It handles the route.

Attensira measures what assistants answer, finds what is holding you back, and opens a pull request you review before anything ships.

Get Started Book A Demo

See who else is in the answer, on your own denominator

In short

Added by domain

No extra runs

Rates that overlap

No invented ranking

The problem it solves

How it works

A worked example

What it does not do

Which plans include it

Frequently asked questions

Does adding a competitor use up my prompt slots?

Why don't competitor rates add up to 100%?

Can I see a competitor's history before I added them?

Do you measure sentiment about competitors?

How should I compare myself to a competitor fairly?

Can I add competitors programmatically?

Keep reading

Mention rate

Citation tracking

Topics

Prompt tracking

You set the destination. It handles the route.

The problem it solves

How it works

A worked example

What it does not do

Which plans include it

Frequently asked questions

Does adding a competitor use up my prompt slots?

Why don't competitor rates add up to 100%?

Can I see a competitor's history before I added them?

Do you measure sentiment about competitors?

How should I compare myself to a competitor fairly?

Can I add competitors programmatically?

Keep reading

Mention rate

Citation tracking

Topics

Prompt tracking

You set the destination. It handles the route.