Best tool for prompt tracking, and the unit arithmetic that decides which one you can use

Prompt tracking means putting a fixed list of buyer questions to the assistants on a schedule and keeping what comes back. Every vendor in and around this category now sells it, and the feature lists all read the same. The thing that actually decides which one you can run your question set in is the unit: what counts as one check, and how many checks your plan holds. Two vendors publish that definition in plain words on their own pages, and when you do the multiplication the same twenty questions turn into very different monthly numbers. This page lists the definitions as each vendor states them, read on 20 October 2026, and shows the arithmetic so you can do it with your own question count.

The criterion this page is ordered on, stated so you can check it

There is no score and no winner here. The ordering is the unit each product meters in, from the narrowest to the broadest, because that is the fact a buyer needs before anything else. We make one of these products and we are in the table on the same terms as everybody else, including the row where our unit is more expensive than theirs. Every figure below was read from the vendor's own live page on 20 October 2026, with HTTP 200 to a Chrome on Windows agent and to a Safari on macOS agent, and the bodies saved.

Read 20 Oct 2026	What one unit is, in the vendor's own terms	Published allowance or cadence	Who writes the questions
Ahrefs Brand Radar	"Each prompt uses 1 check per platform, location and update"	Tiered AI prompt counts, rising from a handful on the lower plans; prompts trackable daily, weekly or monthly; prompts can be organised into tags	Both. A 492 million prompt index modelled from its keyword database, plus a Custom Prompts feature
AI Knows Us, ours	"One credit is one buyer question, asked on one AI app, once. Most checks use one credit. A Claude check uses 3, because Claude costs us more to run"	Plans publish a watched question count of 30, 100, 200 or 300, and an answers read per month figure beside it	You, in a frozen question file, with the brand name absent from every question
BrightLocal Local AI Visibility	A tracked prompt, per location	Up to 20 prompts chosen per location, with suggestions offered from your business category	Both. It suggests from the category and you edit
Meltwater GenAI Lens	A custom prompt run across multiple models	Runs described as every 24 to 48 hours, with weekly runs described elsewhere on the same page	You, branded and unbranded prompts
Brand24 Chatbeat	A tracked prompt across the engines it names	Not stated on the page we read	"You define the prompts you want to track", with suggested prompts offered
SE Ranking	A target query, with the prompts and citations behind the answer shown	Not stated on the page we read; an API and an MCP endpoint are offered for pulling data out	You

The arithmetic, done once so you can do it with your own numbers

Take a modest question set: 20 questions, 4 engines, 1 location, checked weekly. Under the Ahrefs definition, one prompt is one check per platform per location per update, so that month is 20 times 4 times 1 times 4, which is 320 checks. Change the cadence to daily and the same 20 questions become roughly 2,400. Add a second location and it doubles again.

Under our own definition the multiplier sits in a different place. One credit is one question on one app once, so the same 20 questions on 4 apps weekly is also 320, with one exception we publish: a Claude check costs 3 credits rather than 1, because Claude costs us more to run. So if Claude is one of your four apps, that month is 240 plus 240, which is 480 rather than 320. That is our unit being more expensive than the others on this page for one specific engine, and it is stated here because a buyer should find it before the invoice rather than after.

Two conclusions come straight out of the multiplication, and neither depends on anybody's price:

• Cadence costs more than question count. Going from weekly to daily multiplies your consumption by about seven. Adding ten questions adds fifty per cent. Most teams buy more questions and then discover the cadence was the expensive decision.
• Engines are a multiplier, not a feature. Every surface you add multiplies the whole set. Choose the engines your buyers actually use and leave the rest out until there is a reason.

The third limit, which almost nobody publishes

The allowance and the unit are on the pages. The question that decides whether the tool is usable in month four is not: what happens to your history when you edit a prompt. A frozen question file is the only thing that makes two months comparable, so a product that resets a trend line when you reword a question quietly pushes you into never improving the list. Ask it during the trial, in writing, and ask it as two questions: does editing a prompt break the series, and can you keep two versions of the file side by side.

Ahrefs publishes one thing here that helps and is worth copying as a practice whoever you buy from: prompts can be organised into tags, by topic, funnel stage or campaign. A tagged set lets you retire a campaign group without touching the core set that carries your trend.

Where the other vendors beat us on this specific job

• Scale of suggestion. An index of 492 million modelled prompts will propose questions nobody on your team would write. Our question files are hand built.
• Cadence. Meltwater publishes a 24 to 48 hour run cycle and Ahrefs offers daily tracking. Our default is monthly with a weekly round on campaign questions.
• Engine count. Brand24's Chatbeat page names nine surfaces. We name five, and one of our five costs three credits.
• Getting the data out. SE Ranking publishes an API and an MCP endpoint. Our top plan lists a visibility API and we document no MCP endpoint.

Build the question set before you shop

1. Write 20 questions from real enquiries, in the words the buyer used, with your company name in none of them.
2. Give the file a version number and a date, and put it somewhere you own rather than inside a vendor's account.
3. Decide the engines on evidence about your buyers, not on the longest list available.
4. Decide the cadence on what decision the number feeds. A board pack needs monthly. A launch needs weekly for six weeks and then back to monthly.
5. Do the multiplication for each shortlisted vendor, using its own published unit, and only then look at a price.
6. Run the whole set by hand once first, so every quote is measured against a baseline of zero. Sizing that first spend is covered in the smallest AI visibility budget that changes anything.

Questions about prompt tracking allowances and units

What counts as one prompt check?

It depends on the vendor and both of the published definitions are quoted above. Ahrefs counts one per platform, per location and per update. We count one per question per app per run, with Claude at three. Any vendor that will not state its definition in writing has given you your answer.

How many prompts does a business actually need?

Twenty is enough to see a pattern and small enough that one person can read the answers. Plans published in this category run from a handful up to the low hundreds, so the allowance is rarely the binding constraint. The cadence and the engine count usually are.

Should I use the vendor's suggested prompts or my own?

Use the suggestions to find questions you had not thought of, then keep your own list as the one that carries the trend. The suggested set drifts toward the category, and the question that decides a deal usually names a constraint only you know about.

Does tracking more prompts make the number more accurate?

Repeats do more than breadth. Three runs of twenty questions tells you more than one run of sixty, because a single answer is a sample of one and two runs of the same question can disagree completely.

Can I track prompts without buying anything?

Yes, and it is the right first move. Twenty questions, two engines, three runs each is 120 answers, which is an afternoon in the engines' own interfaces with a spreadsheet beside you. A free checker is sized for a narrower job than that, and what each kind is built for is set out in what a free AI visibility scan is sized for. Every quote you receive afterwards has to justify itself against the hand run baseline.

The file, not the dashboard

The thing that gets better over time here is the question file. A dashboard renders it, meters it and charts it, and a different dashboard can render the same file next year. So write the questions down somewhere you own, version them, and judge each vendor on how cheaply it will ask yours rather than on how many it can suggest.