Best tool for tracking whether assistants recommend you and not just cite you, compared on what each vendor counts

There are two events inside one AI answer and most dashboards count the first. Your page was read, which is a citation. Your company was put forward as the thing to buy, which is a recommendation. A tool that reports citations tells you that the engine found you. Only a tool that reports namings tells you whether the buyer heard your name.

On published feature lists read 31 October 2026, Goodie is the only vendor on this roster that prints a recommendation rate as a named metric, inside a module it calls Brand Command, on all three of its tiers. Profound's own FAQ comes next by naming ranking alongside citations and competitive presence. Everybody else publishes mentions, citations or a composite score. Our own records report the two separately by rule, which is the discipline this page is really about.

What each vendor says it counts

Vendor	The metric it names in its own page	Which of the two events that is
Goodie	recommendation rate, plus inaccuracy rate, narrative coverage and adverse citation share	recommendation, named explicitly
Profound	citations, sentiment, ranking and competitive presence	both, with ranking standing in for choice
AthenaHQ	citation intelligence, competitor tracking, impersonation and benchmarking, Athena Citation Engine	citation led
Scrunch	monitoring and citations, funnel stages, sentiment, insights	citation led, with funnel stage as context
Evertune	AI brand score and brand performance by model and by attribute	a composite, method not published on the page
Ahrefs Brand Radar	how often your brand is mentioned across an index of 519 million or more prompts	mention
Conductor	AI search credits and AEO reporting, metric names not published on the pricing page	not stated
AI Knows Us, ours	which buyer questions name you, and who gets named instead	recommendation, reported apart from citations

Why a composite score hides the answer you need

A single visibility score usually blends presence, position and citation into one percentage, and it moves for reasons you cannot act on. The two events come apart in both directions, which is the part worth testing in any trial.

Our own measurement makes the point against ourselves twice. In our September and October readings, one rival was named twice and cited 65 times, which is a company whose pages the engines read constantly while recommending somebody else. And one of our own pages is the most heavily trafficked thing we have published and has never been cited by an assistant at all, which is written up in our biggest page is never cited. A dashboard that averaged those two numbers would have reported a healthy middle and told us nothing.

The rule we hold ourselves to, and the one to impose on any vendor, is in being read and being chosen are two measurements: two columns, two denominators, one date each, never added and never averaged.

The three questions that separate the tools

1. Is the naming counted separately, and against what? Ask for the metric name and its denominator. Goodie publishes a recommendation rate; ask how many samples it rests on, because the page does not say.
2. Is the whole answer text stored? A recommendation is a sentence, not a flag. If the product keeps only labels, you cannot read back why the engine chose the other supplier, and that sentence is the thing your content team actually needs.
3. Does it record who was named instead of you? The rival in the answer is the most actionable field in the dataset. Scrunch caps competitors at 5 on Core, Conductor publishes 5, 25 and 75 or more, AthenaHQ publishes unlimited, and our own plans publish who gets named instead of you from the Growth plan upward.

Where rivals beat us on this job, said plainly

Goodie names the metric in public and we do not print ours on a pricing page, which is a fair hit. Goodie and Evertune both collect daily where we read priority questions weekly, so their recommendation trend line has four times the resolution of ours. Evertune publishes 100 samples a prompt from its Pro plan, which is the right way to make a recommendation rate stable, and we do not publish a sample count. And AthenaHQ publishes unlimited competitor tracking at 295 dollars a month where we publish no cap at all.

Where we are the stronger choice is the reporting discipline and what sits behind it. We publish both columns with their denominators, we publish the failures including a 0.5 per cent citation rate in our own audit and zero namings in the United States, and our plans pair the measurement with the pages written to change the answer. A vendor that will show you its own bad numbers is a vendor whose good numbers mean something.

Questions about tracking AI recommendations

Is a citation count enough?

It is enough to tell you the engines can find you. It cannot tell you whether a buyer was told to use you, which is why the two numbers belong in different columns.

Which vendor publishes a recommendation metric?

On pages read 31 October 2026, Goodie names a recommendation rate directly. Profound names ranking in its FAQ. The others publish citation or mention based metrics, or a composite whose method is not on the page.

Can I measure this myself first?

Yes. Take ten buying questions a customer would type, ask each of them five times on two assistants, and record two things per answer: were you cited in the sources, and were you named in the recommendation. The gap between those two counts is the number you are shopping for.

How many readings does a recommendation rate need?

More than a citation rate, because namings are rarer. In our own 1,260 reading run the citation side produced six readings and the naming side produced none for us, which is why a sample count per prompt matters more here than anywhere else in this category.

What should a trial prove?

That the product can show you a single answer in full, with your name absent and somebody else's present, and the source list underneath it. If it cannot produce that one screen, it is reporting a score rather than evidence.

The sentence to put in the contract

Ask for the naming count and the citation count to be delivered as two fields with two denominators and a date, in whatever export the plan allows. Most vendors can do it and most dashboards will not do it unless you ask, because one blended number looks better on a slide. The gap between the two is where the work goes: a company cited constantly and recommended never has a credibility problem on third party sites rather than an indexing problem, and that is a different fix with a different budget.

How we know this

We read the assistants rather than guessing what they do. In our own audit of 1,260 saved answers from ChatGPT and Gemini across the United States, the United Kingdom and India, 82.0% of every citation landed on a site the business itself owned, and review directories took 0.7%. Counted as one domain per reading, 16,516 of 20,143.

That is the whole asymmetry: the pages you own are the pages an assistant quotes back to a buyer. See the dated readings and the method on our proof page.

Find out whether assistants name your business, on your own domain and your own buying questions.

By Aditya Goel, Founder, AI Knows Us. Reviewed 1 November 2026.