Which AI Visibility Platform Reports the Most Useful Metrics?

Key takeaways

The short version of this page, in the order the sections below develop it. Each point is stated here in full so it can be read without the rest of the page, and the section that argues it gives the measurement and the source behind it.

• Judge a metric by four printed things: numerator, denominator, sample and date. A percentage without them cannot be checked or compared.
• In the documentation we read on October 5, 2026, Profound, Otterly and Scrunch document a calculation for mentions, citations, share of voice, sentiment and position. Peec AI writes out a formula for visibility and share of voice.
• Visibility is not one metric. Peec AI's formula, read literally, divides by total responses; Profound's divides by runs that mention any brand and averages the models with equal weight.
• We have not tested these platforms. A published formula tells you what is counted, not whether the counting is accurate.

Which AI visibility platform reports the most useful metrics?

No platform excels in AI visibility metrics across the board; the useful metrics are the ones whose formula the vendor publishes, and on that test Profound, Otterly and Scrunch document a calculation for all five core metrics. Peec AI writes out a formula for two of the five and defines the rest in words. The Ahrefs and Semrush pages we read describe their metrics in words, without a formula.

We read each vendor's own documentation on October 5, 2026. We have not used these platforms, and nothing here says which one measures more accurately.

What makes an AI visibility metric usable?

A metric is usable when four things are printed next to it: the numerator, the denominator, the sample and the date. Without the first two you cannot tell what was counted; without the last two, how much weight the number bears or whether it is current.

This is the test we apply to our own figures. Mention Rate is the share of a buyer's real questions where an AI assistant names the company. Model Agreement is the share of companies named by at least one assistant that every assistant names.

In our S&P 100 study, Model Agreement was 32 of 48. The numerator is the companies named by all four assistants, the denominator is the companies named by at least one, the sample is 342 usable answers in the eight sectors analysed, and the date is September 2026. ChatGPT returned fewer usable answers than the other three, which pushes that agreement down by an amount we cannot state.

Printing the sample is what exposes that weak point. ChatGPT returned usable text in 54 of 96 answers, against 96 of 96 for each of Claude, Gemini and Perplexity. Fewer ChatGPT answers pushes agreement down, so 32 of 48 likely overstates the disagreement, and we cannot say by how much.

Some of the vendor documentation makes the same point. Scrunch says its charts show the sample size in the tooltip, with the example "43% (37 of 86 responses)". Otterly's sentiment page tells readers to read its sentiment score alongside the count of mentions behind it.

Do platforms calculate the same metric the same way?

Platforms do not calculate the same metric the same way, and the difference is in the denominator rather than the label. Four platforms publish a formula for how often a brand is mentioned, and no two divide by the same thing: one counts responses, another counts only the runs in which some brand was named, discarding the rest from both halves of the fraction. Two products can watch the same answers and print different percentages without either being wrong. The table below sets the published numerators and denominators side by side.

Profound's documentation says runs in which no brand was mentioned are excluded from both the numerator and the denominator. Peec AI's formula, read literally, keeps every response in the denominator. If part of your prompt list produces answers that name no company at all, that exclusion on its own can only raise a score: the same answers give a larger ratio with those runs removed than with them kept.

Averaging is the second trap. Profound averages the models with equal weight, so a model with many runs counts the same as a model with few. Its own worked example shows pooled runs giving 0.53 where the documented method gives 0.50, and the page calls the pooled figure wrong.

Share of voice differs in which brands fill the denominator. Peec AI divides the number of times your brand was mentioned by the total number of brand mentions, which its metrics overview describes as all tracked brands mentioned. The Scrunch metrics reference describes a share across the active competitive set in which every slice adds up to the whole. Profound divides by mentions of all brands and counts each brand once per run.

How do the platforms define sentiment and position?

The platforms define sentiment and position each on its own scale, so neither number carries over from one platform to another. Three vendors publish a sentiment formula and no two agree: one divides positive claims by all claims, one subtracts negative mentions from positive ones, one scores each response, and one maps the result onto a 0 to 100 range. Position splits two ways as well. Treat both as internal to the product that produced them, not as readings of the same underlying thing.

• Profound: positive claims about your brand divided by all claims about your brand, times 100. Every claim is scored positive or negative, so the two values sum to 100.
• Otterly: a Net Sentiment Score, positive mentions minus negative mentions divided by total mentions, times 100. It ranges from −100 to +100.
• Scrunch: each response that mentions the brand is scored 100 for positive, 50 for mixed or 0 for negative, and the metric is the average.
• Peec AI: a 0 to 100 scale based on the language around the mention. The page says most scores fall between 65 and 85 and gives no formula.

Position splits two ways. Profound, Otterly and Peec AI report an average position across the answers in which the brand appears; Profound and Peec AI state that the first brand mentioned is position 1. Peec AI adds that it counts every brand detected in a response, including untracked ones.

Scrunch measures where in the text the brand appears: the top 25% of the response, the middle 50% or the bottom 25%. It reports order of mention separately, as Rank Average. Semrush's Prompt Tracking page describes a third thing, the position of a domain, ranked by where it appears among an answer's citations. Three metrics share one name.

When does this comparison not apply?

This comparison does not apply to accuracy, to anything behind a login, or to any date after October 5, 2026. A published formula tells you what is counted, never how well it is counted. A vendor whose formula is undocumented may still calculate carefully; absence of documentation is not absence of method. Definitions also move, engines drop out of coverage, and at least one vendor notes that editing your own setup can change the history you already have.

• Accuracy. A published formula tells you what is counted. It does not tell you how well brand names are detected.
• Undocumented is not absent. Ahrefs and Semrush may calculate their metrics rigorously. We can only report that the pages we read do not show a formula.
• History can move. Profound's documentation states that editing your configuration changes past numbers, because topics, tags and the brand list are applied at request time. Export a figure on the day you report it.
• Engines drop out. Scrunch notes that Grok no longer contributes citation data. Otterly leaves out AI Overviews that were not triggered. A rate can change because the denominator changed.
• A mention is not a recommendation. Every mention metric above counts being named, not being preferred.

How to check a metric before you buy

Five checks test a metric before you buy, in order, and all five can be done before you pay anything. Ask for the numerator and denominator of every headline rate, for the number of answers behind it, for what gets excluded, and for how several models are combined into one figure. Each question has a correct answer the vendor already knows, so a vague reply is itself a result. Plan limits and prices are a separate question, covered in our guide to tool pricing.

1. Ask for the formula. Request the numerator and denominator of every metric in writing. If the vendor documents them publicly, read that page first.
2. Ask for the sample. Find out how many answers stand behind each rate, per assistant.
3. Ask what is excluded. Answers that name no brand, failed runs and engines that returned nothing all change a denominator.
4. Ask how models are combined. Pooled answers and an equal-weight average of models give different scores.
5. Read the raw answers. A metric you cannot trace to its answers cannot be audited.
6. Run a manual baseline. Check by hand whether ChatGPT mentions your brand on ten questions, then compare what the tool reports for the same ten.

Plan limits and prices are a separate question, covered in our guide to AI visibility tool pricing.

Questions people ask next

These are the follow-ups we are asked most often once the answer above has landed. The same rule applies to each: where a figure appears, its source and year sit in the same sentence, and where we do not know something we say so rather than estimate it.

Is a visibility score from one platform comparable with another's?

No. In the documentation we read on October 5, 2026, Peec AI's formula divides responses that mention your brand by total responses, while Profound's divides by runs that mention any brand and then averages the models with equal weight. Otterly counts prompts and Scrunch counts responses. Compare a score only with earlier readings from the same platform.

Which platforms publish their sentiment formula?

Profound, Otterly and Scrunch do, and each uses a different scale. Profound reports positive claims as a share of all claims, from 0 to 100. Otterly reports a Net Sentiment Score from −100 to +100. Scrunch averages responses scored 100, 50 or 0. Peec AI gives a 0 to 100 scale and no formula on the pages we read.

Does a published formula mean the number is accurate?

No. A formula says what is divided by what. It does not say how reliably the platform detects a brand name, how many times each prompt is run, or whether the answers match what a buyer sees. We have not tested any of these platforms and make no claim about accuracy.

What is the difference between a mention and a citation in these tools?

A mention is your brand named in the text of an answer; a citation is one of your pages used or linked as a source. Peec AI's documentation states that you can be visible as a source without being mentioned as a brand, and the reverse. Every platform in this comparison reports the two separately.

Do Ahrefs and Semrush report share of voice for AI answers?

Ahrefs lists AI Share of Voice among Brand Radar's key metrics. Semrush describes share of voice as part of its Brand Performance reports, and its Prompt Tracking help page says share of voice is not available for the AI engines tracked there. None of the three pages we read on October 5, 2026 gives a formula.

Related

Other questions in this library that cover the ground next to this one. Each is answered the same way — the direct answer first, then the measurement behind it — and each carries its own sources, so you can check any of them on its own page without taking this one on trust.

• SEO Suite or Dedicated AI Visibility Tracker: Which One Should Track Your Brand?
• Which AI Visibility Tool Is Easiest to Start With? What the Free Trials Limit
• AI Visibility Tool Pricing: What Profound, Peec AI and Otterly Charge, and Per What