Guide · 9 min read
GEO KPIs: a five-level framework to measure what AI search brings you
Good GEO KPIs connect what assistants say about your brand with what that does for the business. One number cannot do that. This guide groups the metrics into five levels, from visibility in ChatGPT, Gemini or Perplexity down to leads, explains how to set targets and how often to report, and lists the KPIs that look useful but mislead.

Why GEO needs its own set of KPIs
Classic SEO reporting rests on rankings, impressions and clicks. AI assistants break that chain. A brand can be recommended in hundreds of answers without anyone clicking, and the same question can get a different answer each time it is asked.
So GEO KPIs have to measure presence inside the answer itself, check that what is said is correct, and still follow the visits and leads that do arrive. Each part needs its own source: your question set, GA4 and Search Console, and your CRM.
If you have not set up the measurement yet, start with our method to measure AI visibility. This article assumes you have, or will have, a fixed set of questions run every month.
GEO KPIs by level: the framework
We sort GEO KPIs into five levels. The first four are outcomes, ordered from closest to the answer to closest to revenue. The fifth tracks the work that should move them.
- Visibility: do assistants name you, and how often compared with competitors?
- Quality: when they name you, is the tone favourable and are the facts right?
- Traffic: how many people reach your site from assistants, and what do they do?
- Business: do leads, sales and brand demand grow?
- Activity: is the agreed work being done, month after month?
Visibility usually moves first, traffic later and business results last. If you judge month two by revenue alone, you will stop work that is about to pay off; if you only look at visibility, you never learn whether it matters.
Level 1: visibility KPIs
Visibility metrics come from running the same questions in the same assistants, under the same conditions, every month. Four of them cover most needs:
- Mention rate: answers that name your brand divided by all answers. This is the headline figure.
- Share of voice: your mentions divided by all mentions of the brands you track. It shows your weight against competitors; our guide to AI share of voice covers the calculation and its pitfalls.
- Citation rate: answers that link or cite your domain as a source divided by all answers. It shows whether assistants use your content, separately from whether they know your name.
- Position: where you appear when you are mentioned, counted as 1 for the first brand named, 2 for the second and so on. Report the average only over answers that mention you.
Split every visibility number by assistant and by intent group. Discovery questions such as which software is best for a small agency matter more than questions that already contain your name, where you should appear almost every time.
Level 2: quality KPIs
A mention that says your product was discontinued, or quotes last year's price, counts as a mention and still costs you. Quality metrics stop that from going unnoticed.
- Sentiment: share of mentions that are positive, neutral or negative. Use a written rule and keep it stable.
- Factual accuracy: share of mentions with no errors about your brand. Check price, location, products, opening hours or plans, and who the product is for.
- Open errors: a simple count of distinct false statements found, with the date each was first seen and the source the assistant cited, if any.
Errors are the most actionable finding. They usually trace back to an outdated page on your site or a third-party listing, and fixing that source is often faster than any content project.
Level 3: traffic KPIs
Google Analytics 4 now has an AI Assistant channel in its default channel group. According to Google's channel definitions, sessions get the medium ai-assistant when the referrer matches a list of AI assistants such as ChatGPT, Gemini, Deepseek, Copilot and Grok. Clicks from AI Overviews and AI Mode are excluded and stay in Organic Search.
- AI referral sessions: sessions in the AI Assistant channel, or in your own custom channel group, per month and per assistant.
- Engaged sessions and session key event rate for that channel, compared with Organic Search.
- Landing pages that receive AI traffic: they show which content assistants link to.
Google has not published the full list of referrers it recognises, so we recommend a custom channel group alongside the default one. The step-by-step setup is in our guide to ChatGPT traffic in GA4.
Treat these numbers as a floor. App visits without a referrer and copied links fall into Direct, and an answer that recommends you without a link leaves no trace in GA4 at all.
Level 4: business KPIs
Business GEO KPIs answer the question your finance team will ask. They are the hardest to attribute, so combine several signals.
- Leads and conversions from AI traffic: key events, such as form submissions, demo requests, bookings or purchases, in sessions from the AI channel.
- Self-reported attribution: add an AI assistant option to the how-did-you-hear-about-us field in your forms or sales calls. It catches people who read a recommendation and searched for you later.
- Branded search lift: change in clicks and impressions for branded queries. Search Console's branded queries filter splits the Performance report into branded and non-branded queries for eligible sites, using an AI-assisted classification that Google says can occasionally be wrong.
Branded search moves for many reasons: campaigns, press, seasonality. Read it as supporting evidence next to the visibility trend, never as proof on its own.
Level 5: activity metrics
Activity metrics track the inputs. They do not prove results, but they tell you whether a flat month means the work is not working or the work did not happen.
- Pages optimized or published per month, with the questions each one targets.
- Outside-presence actions completed: directory listings corrected, reviews requested, comparison articles or lists pitched, forum answers posted. Our guide to brand mentions for AI explains which ones tend to matter.
- Errors corrected at the source, linked to the quality KPIs above.
How to set targets for GEO KPIs
Targets set before a baseline are guesses. Run your question set for at least one full month first, ideally twice, and record every level. Then set targets in this order:
- Pick one primary KPI per level, no more. For most brands that is mention rate, factual accuracy, AI sessions with key events, and leads.
- Set visibility targets relative to the baseline and to the strongest competitor, for three and six months. A move from 10% to 15% mention rate on discovery questions is a clear goal.
- Use ranges, not single points. Answers vary between runs, so a target of 14 to 18% is more honest than exactly 16%.
- Make quality targets absolute: zero known factual errors open for more than a month.
- Set business targets as a direction until volumes are large enough to compare. Ten leads a month cannot support a precise percentage goal.
- Agree activity targets from the plan: if the contract says five pages and two outside actions a month, that is the target.
Keep the core question set fixed while targets apply. Adding questions changes the denominator and makes any trend meaningless. New questions go into a separate group, as explained in our guide to prompt research.
Report cadence: what to look at and when
- Monthly: visibility, quality, traffic and activity. Changes in content and outside mentions take weeks to show up in answers, so weekly visibility runs mostly measure noise.
- Quarterly: business KPIs, target review and a decision about the question set and priorities for the next quarter.
- As soon as found: any factual error about prices, safety, legal or medical claims. These are fixed first and do not wait for the report.
A good monthly report fits on one page: the five levels, last month, the baseline, the target range and two or three sentences on what changed and why.
A fictional example dashboard
Fictional example: Tarnwell is an invented project management software company, and every figure below is made up to show the format. It tracks 60 questions in five assistants, run twice a month. Values are month three against the baseline month.
- Visibility. Mention rate 18% (baseline 11%, target 15 to 20%). Share of voice 13% (9%). Citation rate 7% (4%). Average position when mentioned 2.8 (3.4).
- Quality. Positive or neutral mentions 88% (82%). Mentions without factual errors 93% (79%). Open errors 2 (6): an old price on a review site and a discontinued plan.
- Traffic. AI channel sessions 260 a month (140). Session key event rate 4.2% (3.6%).
- Business. Demo requests from AI traffic 11 (5). Three new leads named an AI assistant in the form. Branded clicks up 8%, during a month with a trade show, so the team reads it with caution.
- Activity. Six pages rewritten, three outside-presence actions completed, two errors corrected at the source.
The reading: visibility and accuracy are on track, traffic follows, and business numbers are still too small for firm conclusions. Next quarter keeps the targets and focuses on discovery questions where one competitor still dominates.
GEO KPIs to avoid
These numbers are easy to produce and hard to act on, so we leave them out of GEO KPIs:
- A single AI visibility score from a tool that does not disclose how it is calculated. You cannot explain its changes or check them.
- Screenshots of one answer. One run of one question is an anecdote, whether it shows you first or not at all.
- Total AI sessions without key events. Traffic alone rewards pages that attract the wrong visitors.
- Mentions counted without reading them. A negative or wrong mention is a problem, not a win.
- Activity reported as success. Ten pages published is an input; it becomes a result only if the outcome levels move.
- Search Console impressions presented as AI Overviews data. Google counts AI Overviews and AI Mode inside the Web search totals, with no separate filter.
Putting the framework to work
Start small: one question set, one spreadsheet, the five levels and one primary KPI each. Once the GEO KPIs are stable for a quarter, add detail where decisions need it. The broader planning behind these choices is in our GEO strategy guide.
If you would rather have it done, every RecommendedGEO plan includes regular reporting on these levels, monthly on Starter and more often on Growth and Scale. A free GEO audit gives you the baseline to start from.
Frequently asked questions
Which GEO KPIs should a small business track first?
Mention rate on discovery questions, factual accuracy and leads from AI traffic. Those three cover visibility, quality and business without needing a large team.
How long does it take to see GEO KPIs move?
Visibility often changes within two to three months of consistent work, because assistants need time to pick up new pages and outside mentions. Traffic and business numbers usually follow later and more slowly, which is why targets are set separately for each level.
Can GA4 show all traffic from AI assistants?
No. GA4's AI Assistant channel only sees visits with a recognised referrer. App visits without a referrer and copied links appear as Direct, and clicks from AI Overviews and AI Mode are counted in Organic Search. Treat the figure as a minimum.
Is share of voice better than mention rate?
They answer different questions. Mention rate tells you how often you appear at all; share of voice tells you how much of the conversation you hold against competitors. Report both.
Should GEO KPIs be reported weekly?
Rarely. Answers vary between runs and changes take weeks to appear, so weekly visibility figures mostly show noise. Monthly reporting with a quarterly review of business KPIs and targets works for most brands, plus immediate alerts for serious factual errors.


