A GEO dashboard often shows four or five numbers side by side: visibility, citations, sentiment, share of voice, sometimes position. GEO, or generative engine optimization, is the work of improving how a brand appears in AI-generated answers. In the same week, those numbers can move in different directions. So which one should you act on?
Each metric answers a different question about AI answers, so choose the metric by the decision you are making. None of them is a business outcome. Treat each one as a signal about AI answers, then check your traffic and sales data separately.
For how each vendor calculates them, see our comparison of nine GEO platforms.
What does each GEO metric actually count?
Every GEO metric is a count over a sample of AI answers. A tool sends a list of prompts, meaning questions typed into an AI assistant, and records the answers.
- Mention or visibility rate: the share of sampled answers that name your brand. It answers "Are we named?"
- Recommendation rate: the share of answers that recommend you, not just name you. It answers "Are we endorsed?"
- Position: where your brand appears in the order of brands named. It answers "How prominent are we?"
- Citation: a linked source shown in the answer. It answers "Is our page used as a source?"
- Sentiment: the tone of what answers say about your brand. It answers "How are we described?"
- Share of voice: your brand mentions as a share of all mentions of the brands the tool tracks. It answers "How do we compare?"
Vendors count these out of different totals, such as all answers or only answers that mention you. So a rate with the same name can't be compared across tools.
Each count also depends on the prompt list and on how often each prompt runs, because answers change between runs. A SparkToro blog study (28 January 2026), run with an AI-tracking startup, sent 12 prompts to ChatGPT, Claude and Google's AI 60 to 100 times each. Fewer than once in 1,000 runs did two answers list the same brands in the same order. Its author concluded that visibility measured over many runs is a reasonable metric, but that ranking position means little.
The two metrics people confuse most often are mentions and citations.
Is being cited the same as being mentioned?
No. Two recent studies found that the two often come apart.
A Semrush study of "ghost citations" (9 June 2026) looked at 3,981 domain appearances across 115 prompts. Semrush sells SEO (search engine optimization) software. In the study, 61.7% of appearances were a cited domain whose brand the answer never named. Another 25.1% were brands named without a citation, and only 13.2% were both.
A BuzzStream study (updated 14 July 2026) analysed 12,000 responses about roughly 200 brands. BuzzStream sells PR outreach software. Only 23.1% of brand mentions included a citation of that brand's own site. For list or category prompts, the figure fell to 7.2%.
The two studies count differently, so don't compare the percentages.
What this means (our reading, not a tested finding): the two gaps call for different work.
- Cited but not named. Your content is being used, but your brand is not credited. Check how clearly the cited pages identify your brand.
- Named but not cited. Other sources are speaking for you. Look at which third-party domains the engines cite for your topics. That is PR, review and off-site work.
An Ahrefs study (May 2025) of 75,000 brands is consistent with looking off-site. Ahrefs sells SEO tools. Of the factors it tested, branded web mentions had the strongest correlation with brand visibility in Google's AI Overviews, the AI summaries shown above search results. Ahrefs itself calls its correlations "moderate to very weak" and notes that correlation is not causation.
Which engines show citations at all is covered in our engine coverage guide. Being named, though, is only half the picture. How you are described is the other half.
What can a sentiment score tell you?
Less than it looks. Vendors use different scales, as documented on 10 October 2026:
- Otterly.AI reports a net score from −100 to +100.
- Peec AI uses 0 to 100 and says most scores fall between 65 and 85.
- Profound averages the positive share across AI models.
- Scrunch shows the share of prompts with positive, mixed or negative answers, according to a July 2026 company blog post, not product documentation.
None of the vendors we checked names the model or method that labels tone, so scores can't be compared across tools.
Scores may also rest on few answers. One July 2026 preprint by Żatuchin found that most answers carried no measurable tone. In it, 91.9% of 12,933 brand answers scored exactly neutral. The scorer was an open classifier trained on Twitter posts, not any vendor's tool. The author concluded that "a single response carries almost no brand-discriminating signal."
When most answers are neutral, a score moves on the handful that are not, so read those answers rather than the number.
Tone is also not accuracy. A warm answer can still state the wrong price. Profound, Peec AI and Genezio each document a separate feature that compares AI claims with facts the brand supplies.
For positioning work, use the statements and an accuracy check, and treat the score only as a trend within one tool. The next metric, share of voice, is the one most often presented as the competitive key performance indicator.
Is AI share of voice the share of voice marketers already know?
It has the same name but measures something different.
In the advertising rule marketers usually cite, share of voice means a brand's share of category ad spend. Les Binet and Peter Field studied the advertising-effectiveness databank of the IPA, the UK's Institute of Practitioners in Advertising. Their best-known rule: brands whose share of voice exceeds their market share tend to grow, all other factors being equal.
A 2020 summary from LinkedIn's B2B Institute puts the effect in business-to-business markets at about 0.7% market-share growth a year. That is per 10 points of share of voice above market share. The figure comes from a report LinkedIn's B2B Institute commissioned from Binet and Field. LinkedIn sells ads.
GEO share of voice is your share of brand mentions in sampled AI answers. Ad spend is an input you control. Mentions are an output, counted against a competitor list that you or the tool picked. Change that list and the number changes.
AI share of voice is also zero-sum. Genezio's documentation notes that your share can fall while your visibility rises, if a competitor grows faster. Vendors also calculate it in different ways, as the formula table in our comparison shows.
We found no published study that validates GEO share of voice against market share. AthenaHQ, which sells GEO software, nonetheless applies the old growth rule to AI share of voice. That transfer is a vendor claim, not evidence.
Use AI share of voice for competitive reporting and for spotting new entrants, with a fixed competitor list, not as a growth forecast. If none of these metrics predicts growth, what is the link to traffic and sales?
Does more AI visibility mean more traffic or sales?
Not necessarily. The evidence so far shows associations, not cause and effect: two things moving together, without proof that one drives the other.
Seer Interactive, a marketing agency, states the problem in its April 2026 study of its clients' data. Brands cited in Google's AI Overviews got more organic clicks per impression than uncited brands. But, as Seer writes, "We cannot claim causation. Higher-authority brands are also more likely to be cited." In other words, the brands that get cited may already be the ones that get clicked. The sample is also concentrated: one client account made up 47% of the uncited impressions for informational queries.
Independent data also shows how little traffic AI answers send directly. Pew Research Center is an independent research organization that sells no marketing tools. It studied US browsing data from March 2025 and published it in July 2025. Google users clicked a traditional result in 8% of visits with an AI summary, against 15% without one. Only about 1% of visits with a summary included a click on a link inside it. Pew reports these as differences, not as proof that the summaries caused them.
AI traffic is also partly hidden in analytics. In Google Analytics 4 (GA4), the default channels sort these visits in three ways:
- Clicks from ChatGPT, Gemini and similar assistants fall into an AI Assistant channel.
- Clicks from Google's AI Overviews and AI Mode count as Organic Search.
- Sessions with no information about where the visitor came from are processed as direct traffic.
So report visibility next to outcome data: the GA4 AI Assistant channel, Search Console, branded search and conversions. A visibility change is a reason to check that data, not a substitute for it.
Which metric should you watch for which decision?
| Metric | What it counts | What it can tell you | What it can't establish | Decision it supports |
|---|---|---|---|---|
| Mention or visibility rate | Share of sampled answers naming you | Whether you're in the answer for your prompts | Whether anyone saw, chose or clicked | Which topics and prompts to prioritize |
| Recommendation rate | Answers that recommend you | Whether presence becomes endorsement | Real buyer preference | Positioning and differentiation |
| Position | Your order among brands named | Rough prominence at best, averaged over many runs | Anything from a single answer, and little even on average | A secondary check only |
| Citation of your pages | Answers linking your site | That your content is used as a source | That your brand is named or clicked | Content updates and technical access |
| Cited third-party sources | Domains engines link for your topics | Where engines get their information | That coverage there will earn a mention | PR, reviews and off-site outreach |
| Sentiment | Tone of answers about you | Direction of tone within one tool | Accuracy; comparison across tools | Messaging fixes, after reading statements |
| Share of voice | Your mentions ÷ all tracked brand mentions | Relative presence against chosen rivals | Market share or growth | Competitive reporting |
| Outcome data (not a GEO metric) | GA4 sessions, branded search, conversions | Visits and conversions you can attribute | AI influence from answers that never send a visit | Budget and channel decisions |
Only the last row is a business outcome. As of 10 October 2026, we found no controlled study linking AI visibility, citations, sentiment or AI share of voice to revenue or market share.
Pick the metric that matches the decision in front of you, and compare its trend within one tool. Change one thing at a time. Then look for movement in your own traffic and sales data, without assuming the visibility change caused it.