AI Visibility Tools and Metrics: How to Measure Whether AI Assistants Mention Your Brand
AI visibility tools measure whether, how often and how favorably AI assistants such as ChatGPT, Gemini, Perplexity, Claude and Google’s AI Overviews and AI Mode mention or cite your brand when people ask questions in your category. They work by running a set of prompts on each platform on a schedule, parsing the answers, and reporting metrics such as mention share, citation share, share of voice, average position and sentiment against competitors. The same measurement can be done by hand with a spreadsheet and a fixed prompt list, which is where most brands should start, because the discipline is in the prompt set and the cadence, not the software. The market now includes modules inside Semrush and Ahrefs, specialist platforms such as Profound, Peec AI, Scrunch and Otterly, and a long tail of smaller tools. This guide covers the metrics worth tracking, how to build the prompt set, the manual method, what each named tool tracks according to its own product pages, and how to attribute a change in the numbers to the press or content that caused it.
Key takeaways
- Six metrics cover what matters: mention share, citation share, share of voice, average position, sentiment and accuracy, measured across a fixed set of prompts and compared with named competitors.
- Individual AI answers vary run to run, so every metric is a sample. Track share across 30 to 50 prompts and judge direction over three monthly runs, not any single answer.
- A spreadsheet and two hours a month will produce a defensible baseline for most brands. Buy a tool when you need daily cadence, many prompts, many markets or several brands.
- Tools differ in whether they track citations (your URL linked) as well as mentions (your name in the text), and in which platforms they cover; check both before paying.
- Attribution works by dating every press placement, profile change and content release on the same timeline as the prompt runs, and reading which cited sources changed.
What should you measure?
There is no rank to track in a generated answer, so AI visibility borrows from brand measurement. The metrics below are the ones that a tool, or a careful spreadsheet, can produce consistently.
| Metric | Definition | What it tells you | How to compute it |
|---|---|---|---|
| Mention share | Share of prompts in your set where the answer names your brand | Whether you exist to the model in this category | Prompts with a mention divided by total prompts, per platform |
| Citation share | Share of prompts where the answer links to a page on your domain | Whether your content is a source, not just a name | Prompts citing your domain divided by total prompts |
| Share of voice | Your mentions as a share of all brand mentions in the prompt set | Your position relative to competitors | Your mentions divided by the sum of mentions across tracked brands |
| Average position | Where your brand appears when several are listed | Whether you are the lead recommendation or an afterthought | Mean of your rank in each answer where you appear |
| Sentiment and framing | Whether you are recommended, mentioned neutrally or flagged with a caution | Whether being mentioned is helping | Classify each mention; report the share that is positive |
| Accuracy | Whether the facts stated about you are correct and current | Which sources are feeding the model wrong information | Check each statement against your canonical facts; log the error and its likely source |
| Prompt coverage | Share of your category’s buying questions for which any answer names you | Where the gaps are by topic and funnel stage | Tag prompts by topic and stage; compute mention share per tag |
Two of these deserve emphasis. Citation share is different from mention share, and many brands have one without the other: cited for a definition but never named as an option, or named as an option on the strength of third-party sources while their own pages are never linked. And accuracy is the metric that most often reveals a reputational problem, because the model is summarizing whatever it found, including old press, stale directories and hostile threads. How to audit that directly is in what ChatGPT says about your company.
How to build the prompt set
The prompt set is the measurement instrument. A bad one produces numbers that move for no reason; a good one is stable enough to show real change. Build it once and change it rarely.
- Start with real questions, not keywords. People ask assistants in sentences: “which PR agency should a Series A startup hire,” not “pr agency startup.” Pull from sales calls, support tickets, People Also Ask boxes and the questions your team hears at events.
- Cover the funnel. Informational (“how does guaranteed press placement work”), comparison (“earned media vs paid placements”), commercial (“best reputation management firm for a doctor”) and branded (“is [brand] legitimate,” “who runs [brand]”). Aim for 30 to 50 prompts, with at least a third commercial.
- Tag every prompt. Topic, funnel stage, branded or unbranded, and the competitors you expect to appear. The tags are what let you say “we gained in comparison prompts but not commercial ones.”
- Write prompts the way a stranger would. No brand name in unbranded prompts, no leading phrasing. If a prompt only produces your brand because it contains your brand, it measures nothing.
- Fix the platforms and settings. ChatGPT with search enabled, Gemini, Perplexity, Claude with web search, Google AI Mode and AI Overviews where they appear. Logged out or in a clean account, same country, same device type each run. Memory and personalization distort results; turn them off for measurement accounts.
- Keep a change log. If you add or reword a prompt, date it and keep the old version’s history separate. Trend lines are only meaningful on an unchanged set.
Ahrefs reports that its Brand Radar expands keyword-database queries into natural questions using People Also Ask and semantic fan-out, which is a useful model for anyone building a set by hand: start from the queries you already rank for, turn each into the three or four questions a person would ask an assistant, and keep the ones that read naturally.
The manual method: a spreadsheet and two hours a month
Before paying for software, run this for three months. It produces a baseline you own and teaches you what to look for in a tool.
The sheet
One row per prompt per platform per run. Columns: date, platform, prompt, tags, brand mentioned (yes/no), position (if listed), our URL cited (yes/no and which), competitors mentioned (list), cited domains (list), sentiment (positive/neutral/caution), factual errors (free text), and a link to the saved answer. Save every answer as a screenshot or exported text; you will want to re-read them.
The run
Once a month, on the same week, run every prompt on every platform in a clean session. Thirty prompts on five platforms is 150 answers, which takes about two hours if you paste and tag as you go. Do not re-run a prompt to get a better answer; the first answer is the data point.
The report
A pivot table gives you mention share, citation share and share of voice by platform and by tag. Add a short written section: the three most common cited domains in your category, the errors found, and the competitors gaining. The cited-domains list is the most actionable output, because it tells you where the model is getting its information and therefore where you need to be present. The reasons those domains are chosen are explained in how ChatGPT and Perplexity choose their sources.
When to upgrade
Move to a tool when the manual run exceeds what one person can do in a morning: more than about 50 prompts, weekly or daily cadence, several countries, several brands, or a need to see historical data you did not collect yourself. The cost of a tool is small next to the cost of a decision made on a sample of ten answers.
What the AI visibility tools track
The tools below are described from their own product pages or help documentation at the time of writing. Coverage and features change often; verify before buying. Prices are given only where the vendor’s page showed them.
| Tool | Platforms covered (per vendor) | Metrics tracked (per vendor) | Prompt model and cadence | Notes |
|---|---|---|---|---|
| Semrush AI Visibility Toolkit | Google AI Overviews, AI Mode, ChatGPT, Perplexity, Gemini (Brand Performance) | Visibility baseline, share of voice, sentiment, mentions and citations vs up to four competitors, most-cited pages, how AI describes the brand | Prompt Research from Semrush’s prompt database; Position Tracking of custom prompts daily | Free plan with basic mentions, citations and visibility score; toolkit from $99/month; Semrush One (with SEO) from $199/month |
| Ahrefs Brand Radar | AI Overviews, AI Mode, ChatGPT, Perplexity, Gemini, Copilot; Claude for custom prompts only | Mentions, citations, AI share of voice, estimated impressions weighted by search volume | AI Visibility Index: prebuilt database re-tested monthly with history back to 2025; Custom Prompts: your own questions daily, weekly or monthly | Custom Prompts from €47/month and free with paid Ahrefs plans; AI Visibility Index from €179/month |
| Profound | ChatGPT, Perplexity, Claude, Gemini, Copilot, DeepSeek, Google AI Overviews and others | Visibility rank and score by platform, citation analytics, sentiment and the sources driving it, FactCheck for inaccurate claims, agent analytics (how bots view and cite your pages), prompt volumes, ChatGPT Shopping product visibility | Prompt volumes drawn from real user conversations; benchmarking against peers | No pricing shown on the product page; enterprise orientation |
| Peec AI | ChatGPT, Perplexity, Gemini, Google AI Mode and others per model filter | Visibility (share of chats mentioning you), average position, sentiment score, sources split into “used” and “cited,” competitor quadrant | Your own prompts with tags for persona and funnel stage; AI-suggested prompts with volumes; each prompt runs once every 24 hours per model | No pricing shown; free trial |
| Scrunch | “Major LLMs”; platforms not itemized on the page | Presence and position by prompt, topic and entity; share of voice; citation analysis; competitive benchmarking by persona, geography and segment; real-time AI bot crawl feed and crawl-error detection; content-gap diagnostics; alerts | Prompt and topic tracking with scheduled reports | No pricing shown; 7-day trial. Sentiment and accuracy not listed on the page |
| Otterly.AI | ChatGPT, Google AI Overviews, AI Mode, Perplexity, Copilot, Gemini, Claude | Brand mentions, average position, coverage over time, link citations, share of AI voice, sentiment, competitor alerts, results by engine and country | Prompt library with filters for funnel stage and branded vs unbranded | Starting price shown as $29/month; trial length shown inconsistently as 7 or 14 days |
Three practical distinctions when comparing them. First, does the tool run your prompts or its own? A prebuilt index (Ahrefs’ Visibility Index, Semrush’s prompt database, Profound’s prompt volumes) gives instant history and category context but may not contain the questions your buyers ask; custom prompts give precision but no history before you start. Second, does it distinguish mentions from citations? Peec’s “used” versus “cited” split and Ahrefs’ separate mention and citation counts are the kind of detail that makes the data actionable. Third, does it show the cited sources, not just your score? A tool that tells you your visibility fell 6 points is less useful than one that shows which domains replaced you.
Cadence: how often to measure and how to read the numbers
AI answers are not deterministic. The same prompt on the same platform an hour apart can list five brands and then four. This is why every metric here is a share across a prompt set rather than a result for a prompt, and why cadence matters.
- Monthly is enough for most brands measuring by hand. It matches the speed at which new coverage and profile changes propagate into answers, and it keeps the sample consistent.
- Weekly suits brands with active press or content programs that want to see effects land, and anyone tracking a reputational issue.
- Daily is a tool-only cadence, useful for large prompt sets where day-to-day noise averages out, and for alerting when a competitor or a negative source suddenly appears.
Read direction, not points. A move from 40 to 46 percent mention share on 30 prompts is two extra prompts and may be noise; the same move sustained over three runs, concentrated in one tag, with a new cited domain appearing alongside it, is a signal. Report platform by platform, because the engines differ: Semrush’s 2025 study of 230,000 prompts found ChatGPT’s citation mix shifted sharply within six weeks while Perplexity’s and Google AI Mode’s stayed comparatively stable, so a drop on one platform with no change on the others is usually that platform, not you.
How to attribute changes to press, content or profile fixes
The question executives ask is whether the press placement or the content program moved the number. With a disciplined timeline, you can usually answer it.
- Date everything on one timeline. Every placement going live, every profile correction (LinkedIn, Crunchbase, Wikidata, Google Business Profile), every page published or rewritten, and every prompt run. One sheet, one date column.
- Watch the cited domains first. If a placement in a recognized outlet is the cause, that outlet’s domain will start appearing in the citations for the prompts where your mention share rose. If a rewritten page is the cause, your own domain’s citation share rises on the prompts that page answers. If a profile fix is the cause, accuracy errors disappear and the sentiment classification shifts.
- Use the tags. A placement about your founder’s expertise should move branded and expertise prompts, not every prompt. If the lift is where you expected it, the attribution is plausible; if it is everywhere at once, suspect a platform change.
- Check the other platforms. A change on all engines at once that lines up with a dated action is strong evidence. A change on one engine that lines up with nothing you did is usually that engine.
- Allow for lag. Live-searching engines can reflect a new indexed article within days; changes that depend on the engine trusting a new source, or on training-data refreshes, take one to three months. Judge each action over a quarter.
This is also how to evaluate whether coverage was worth buying. The mechanism by which placements turn into recommendations is explained in how press coverage gets your brand recommended by ChatGPT, and what to do with a feature once it runs is in getting ROI from a media placement. The relative weight of mentions and links in this picture is covered in brand mentions vs backlinks for AI answers.
Common measurement mistakes
Measuring from a personal account. ChatGPT’s memory and Google’s personalization shape answers. Use clean accounts with personalization off.
Changing the prompt set every month. New prompts reset the trend. Add in batches, keep the originals, and report the original set separately.
Counting a mention as a win. A mention that says “some users report billing problems with [brand]” is a reputational finding, not a visibility gain. Sentiment and accuracy belong in every report.
Ignoring citations. If competitors’ pages are cited and yours are not, you have a content extractability problem that no press placement will fix; the on-page checklist in what generative engine optimization is addresses it.
Treating a tool’s score as the metric. Proprietary visibility scores are convenient for dashboards but opaque. Report the underlying shares, which you can explain to a board and reproduce by hand.
Forgetting Google. AI Overviews and AI Mode sit on the search page most buyers still use. Include them, and read our guide to ranking in Google AI Overviews and AI Mode for why their citations behave differently.
Where Viral Spike fits
Viral Spike Marketing’s AI search visibility service begins with exactly the measurement described here: a prompt set built from your category’s real buying and reputational questions, run across ChatGPT, Gemini, Perplexity, Claude and Google’s AI features, with mention share, citation share, sentiment, accuracy and the cited-domain list reported platform by platform. The same set is re-run monthly through the engagement so that every placement, profile fix and content change is attributed on one timeline. The corroboration layer uses guaranteed editorial placements, where you approve the draft, the article is published and indexed on Google, and the engagement is refunded in full if it never runs. If you would rather see the baseline before committing to anything, contact us.
Build the set, run it once, and keep the answers
Write 30 prompts this week: ten informational, ten commercial, five comparison, five about your brand by name. Run them on three platforms in a clean session, save every answer, and fill the sheet. You will finish with three numbers (mention share, citation share, share of voice), a list of the domains the engines are using in your category, and a list of errors. That is the baseline every later decision is measured against, and it is the only way to know whether the next press placement or content program did anything.
Frequently asked questions
What is an AI visibility tool?
An AI visibility tool runs a set of prompts on AI platforms such as ChatGPT, Gemini, Perplexity, Claude and Google’s AI Overviews and AI Mode on a schedule, parses the answers, and reports how often your brand is mentioned or cited, where it appears, how it is framed and how that compares with competitors. Examples include modules inside Semrush and Ahrefs and specialist platforms such as Profound, Peec AI, Scrunch and Otterly.
What is the difference between a mention and a citation in AI answers?
A mention is your brand name appearing in the answer text. A citation is a link to a page on your domain as a source for a claim. They are different metrics: a brand can be named on the strength of third-party sources while its own pages are never linked, or cited for a definition without being named as a recommended option. Track both, and report share across a fixed prompt set rather than single answers.
How many prompts do you need to measure AI visibility?
Thirty to fifty is a practical minimum for a single brand in one market: enough to spread across informational, comparison, commercial and branded questions and to make a share figure meaningful, while still being runnable by hand in a couple of hours a month. Larger sets, several markets or daily cadence are the point at which a tracking tool pays for itself.
Can you measure AI visibility without paying for a tool?
Yes. Build a fixed prompt set, run it monthly on each platform in a clean session with personalization off, and log in a spreadsheet whether you were mentioned, your position, whether your URL was cited, which competitors and domains appeared, sentiment and any factual errors. A pivot table gives mention share, citation share and share of voice. Most brands should do this for three months before buying software.
How do you know if press coverage improved AI visibility?
Date every placement, profile change and content release on the same timeline as your prompt runs, then look at which cited domains changed. If a new outlet appears in the citations for the prompts where your mention share rose, and the lift is concentrated in the topics the article covered, the attribution is sound. Allow one to three months for engines to trust a new source, and judge over a quarter.
How often should AI visibility be measured?
Monthly is enough for most brands measuring by hand, because it matches how quickly new coverage and profile changes show up in answers and keeps the sample consistent. Weekly suits active press programs or a reputational issue in progress. Daily is a tool-only cadence for large prompt sets and alerting. In every case read direction over several runs, since individual AI answers vary.
Sources
Primary sources referenced in this article, in order of first use.
- Semrush AI Visibility Toolkit (semrush.com)
- Ahrefs Brand Radar (ahrefs.com)
- Profound (tryprofound.com)
- Peec AI (peec.ai)
- Scrunch (scrunch.com)
- Otterly.AI (otterly.ai)
More in AI Search
