Why AI visibility needs its own KPIs
Buyers now build their shortlist inside ChatGPT, Gemini, Perplexity and Google's AI Overviews. A brand can rank first on Google and still be missing from the AI answer. Being mentioned is not the same as being recommended.
Classic SEO metrics (rank, clicks, impressions) miss most of this. Much AI influence leaves no click: a buyer reads an answer, remembers a name and types your URL a week later.
The CMO KPI set
Tier 1: Are we in the answer?
|
KPI |
What it tells you |
How it's calculated |
Cadence |
|---|---|---|---|
|
Visibility |
How often AI names you |
Answers that mention you ÷ all answers, on a fixed prompt set |
Weekly |
|
Prompt coverage |
How many of your key questions you show up in at all |
Prompts with at least one mention ÷ all key prompts |
Monthly |
|
Share of voice |
Your visibility against named competitors |
Your visibility vs. each tracked competitor's |
Monthly |
Tier 2: Does AI choose us, and what does it say?
|
KPI |
What it tells you |
How it's calculated |
Cadence |
|---|---|---|---|
|
Recommended rate |
How often AI picks you over the alternatives |
Answers that recommend you ÷ all non-branded answers. A mention or a #1 listing doesn't count |
Weekly |
|
Sentiment |
How AI describes you |
Positive ÷ (positive + negative) statements about your brand |
Monthly |
|
Citation share |
How often AI uses your own pages as a source |
Citations to your site ÷ all citations in your answers |
Monthly |
|
Accuracy |
Whether AI gets your facts right (prices, markets, features) |
Answers with a factual error ÷ answers that mention you |
Monthly |
Tier 3: Does it move the business?
|
KPI |
What it tells you |
How it's calculated |
Cadence |
|---|---|---|---|
|
AI-referred sessions |
Visits that arrive from an AI assistant |
Sessions from AI referrers in your analytics |
Weekly |
|
AI-sourced pipeline |
Leads and deals from those visits, plus buyers who say AI sent them |
CRM source field + a "How did you hear about us?" field |
Monthly |
|
Intervention lift |
Whether a specific action moved your visibility |
Change on the prompts you targeted, minus the change on prompts you didn't touch |
Per action |
If the board gets three numbers: recommended rate on your must-win questions, AI-sourced pipeline, and the lift from what you shipped.
Which tool measures which link
|
|
3RD |
Peec AI |
Profound |
GA4 |
CRM + "how did you hear" field |
|---|---|---|---|---|---|
|
Visibility and competitor benchmark |
Yes |
Yes |
Yes |
No |
No |
|
Recommended rate (chosen, not just mentioned) |
Yes |
Not stated (tracks visibility and position) |
Not stated |
No |
No |
|
Citations and sources |
Yes, owned vs. earned |
Yes |
Yes |
No |
No |
|
Sentiment |
Yes |
Yes |
Yes |
No |
No |
|
Accuracy of AI claims |
Partly (flags outdated claims and their sources) |
Not stated |
Yes (FactCheck) |
No |
No |
|
AI-referred sessions, conversions, revenue |
Yes, from your GA4 (Growth and Enterprise) |
Not stated |
Partly (Agent analytics) |
Yes, with an AI channel group |
No |
|
Leads and deals from AI |
No |
No |
No |
No |
Yes |
|
Lift per action, against a control |
Yes |
No |
Not stated |
No |
No |
|
Helps execute the fixes |
Yes: content drafts, outreach, crawlability tasks |
Strategy recommendations |
Yes: content, page and correction drafts |
No |
No |
The Peec AI and Profound columns are based on their own websites as of November 2026. If something has changed, tell us and we'll update it.
What each option gives you
- 3RD measures whether AI recommends you, not only whether it names you, and splits citations into your pages and other sites. With GA4 connected (Growth and Enterprise), AI-referred sessions, conversions and revenue sit next to visibility. Every completed task is measured against your untouched prompts. 3RD doesn't tie a deal to an answer; that stays in your CRM.
- Peec AI is a clean, fast tracker for visibility, position and sentiment, with source analysis, a Looker Studio connector, an API and an MCP server. Its site doesn't list a recommendation metric, analytics connection or measured effect per action.
- Profound is strongest on accuracy (FactCheck) and on how AI bots read your site (Agent analytics). Its AI Marketer drafts content, page fixes and corrections for approval. Its site doesn't describe a lift measurement against a control.
- GA4 counts the AI visits that click through, once you add an AI channel group. Some assistants strip the referrer, and AI Overviews clicks arrive as normal Google traffic.
- Your CRM plus a "How did you hear about us?" field is the only place pipeline lives, and the only direct evidence of zero-click influence.
When to choose which
- You only need to know if AI mentions you: a lightweight tracker such as Peec AI is enough.
- Wrong facts about you in AI answers are the main risk (regulated, technical or enterprise products): Profound's FactCheck is built for that.
- You must show the board what AI visibility work delivered, across several markets and languages: 3RD, connected to GA4 and read next to your CRM.
- Whatever you choose: add the "How did you hear about us?" field this week. It costs nothing and catches what no tool sees.
What this means: no single tool covers the whole chain. A visibility platform covers the top half. Your analytics and CRM cover the bottom half. Report them side by side, never blended into one revenue number.
Observed, attributed, inferred: keep them apart
- Observed: AI named, recommended or cited you. Measured directly.
- Attributed: a visit arrived from an AI assistant and converted. Measured in analytics and the CRM.
- Inferred: a buyer saw you in an AI answer, then came back through a branded search or by typing your URL. This can only be estimated, for example with a "How did you hear about us?" field.
Most AI influence sits in the third layer. Don't give the board a revenue figure that pretends otherwise. Present attributed revenue as a floor, not the total.
Intervention lift: the part you can prove today
You can't trace one answer to one invoice. You can prove that your work moved the answers.
- Fix a set of prompts and keep it constant. Changing prompts mid-test breaks the comparison.
- Record a dated baseline: visibility, recommended rate and citation share.
- Ship one action: a new page, an outreach placement, a technical fix. Note the prompts it targets.
- After a fixed window, compare the targeted prompts with the ones you didn't touch.
- Report the difference as that action's lift. Then check analytics and the CRM for matching AI visits and leads.
In 3RD, every completed task shows its measured effect next to your other prompts, for example "Visibility 29% → 17% · Other prompts +4%". That example shows a fall, and that is the point: an honest measure reports what happened, not what you hoped for.
Why your own website isn't enough
Across 100 brands and 5.9 million AI citations (19 September to 18 October 2026), the median brand's own website made up just 6.8% of the sources AI cited. A re-run a month later across 364 brands and 15.5 million citations gave the same 6.8%. The rest came from other sites: reviews, comparisons, forums, media and competitors.
If your KPIs only track your own site, they miss most of what shapes the answer. Track earned citations too.
Example: GetWhy
GetWhy, an AI research platform, used the questions it was losing as its content plan. Between the first 30 days (24 April to 24 May 2026) and the latest 30 (27 September to 26 October), its AI visibility rose from 1% to 14%, and from 1% to 17% on must-win questions.
The business number came from a plain form field. Leads who named an AI assistant when asked how they found GetWhy rose from 4% of inbound in May–August to 16% in September–October.
That is the full chain on one page: visibility measured in 3RD, and leads measured in the CRM. Read the GetWhy case.