The 3RD glossary: 15 metrics behind your dashboard

Your 3RD dashboard shows 15 numbers. Two of them tell you most of what you need: Visibility on your Must-Win prompts, and Recommended rate right next to it. The median brand we track is mentioned in 19.8% of AI answers and recommended over the alternatives in 0.2%. If you only read the first number, you'll think you're on a shortlist you're actually losing.

AI visibility rests on six core metrics: Visibility, Recommended rate, Citation Share, Sentiment, Position and Rank. Here is each one, defined and explained.

Most GEO (generative engine optimization) glossaries explain the industry's shared vocabulary: GEO, AEO, citations, hallucinations. This one does a different job. It defines the 15 metrics you'll actually read inside 3RD, in one quotable sentence each. Then it shows how each is calculated, what good looks like, and what to do when the number drops.

It follows the order you use them in: first how you're doing (Track), then why (Diagnose), then what to do about it (Act). The last section covers which of these numbers a CMO should report, how often, and on one page.

For general GEO terms, see the GEO glossary: 15 terms every marketer should know. This page covers the metrics.

Two things make this glossary different. First, the running example is a demo workspace: a sportswear brand tracked against six competitors in Denmark, Sweden, the UK and the US, across seven AI models and 207 prompts. It isn't a customer's account, but its numbers are real, and they show how the metrics read together. Second, for the core metrics we show what "normal" looks like across 111 brands tracked in 3RD, so you can tell a bad number from a typical one.

Enlarge

The Overview for the past week: 44.5% visibility, mentioned on 172 of 207 prompts, split by prompt tier. Demo workspace; brand names blurred.

Updated October 19, 2026. Definitions match the 3RD platform and docs.get3rd.com.

The 15 metrics at a glance

The six core metrics

These six answer the first question every marketing team asks: how are we doing in AI answers?

1. Visibility

Visibility is the share of AI answers that mention your brand. It's the headline number, and it answers the most basic question: are you in the conversation when buyers ask AI about your category?

How it's calculated:

Visibility = (answers that mention your brand ÷ total answers) × 100

Each prompt is scored on its own, and the overall score is the average across prompts. Every prompt counts equally, however many AI models answered it. Branded prompts, the ones that name your brand, are left out of visibility scores by default, because they inflate the number.

What good looks like: 3RD recommends a target per prompt tier: 75% on Must-Win prompts, 50% on Core and 35% on Foundational (see #7). On competitive "best X" prompts, the real goal is a rising number at or above your nearest competitor. Don't chase 100%.

When it drops: check which filter moved first. Visibility can be sliced by model, market, topic and audience, and a drop usually lives in one of them. Then read the answers behind it in Conversations: did a competitor start appearing, or did a new source get cited?

Example: a snack brand tracking 90 prompts has an overall Visibility of 42%. On its Must-Win prompts alone it sits far lower. It shows up in general conversation but disappears on the questions closest to a purchase. That second number is the one to act on.

Visibility over time is the Visibility development chart on the Overview. Trend direction tells you more than any single day's number.

What normal looks like: across 111 brands with a stable daily series (and at least 1% visibility), the median brand's daily visibility moved 12.5 points between its best and worst day of the month, with no change to its prompts. A 5-point drop from one day to the next is usually noise. A 5-point drop in the 30-day trend is not.

2. Recommended rate

Recommended rate is the share of AI answers that recommend your brand over the alternatives. Visibility tells you whether you're in the answer. Recommended rate tells you whether the answer tells the buyer to pick you.

What counts: an answer recommends a brand when it clearly prefers it, for example by calling it the best option or telling the reader to go with it. Being mentioned doesn't count. Being listed first doesn't count. Being called "the market leader" doesn't count either: that describes a brand, it doesn't recommend it.

How it's calculated:

Recommended rate = (answers that recommend your brand ÷ answers checked) × 100

Branded prompts are left out. Each prompt is scored on its own, and the overall rate is the average across prompts. 3RD has counted it since 22 September 2026.

A related number, Recommendation share, answers a narrower question. Of the answers that pick any winner, how often is the winner you or one of your products?

What good looks like: compare your rate to your competitors', not to 100%. Many prompts ask for a list, not a choice, and many answers decline to pick ("it depends on your needs").

What normal looks like: low. Across 60 brands and 494,000 checked answers, the median brand was mentioned in 19.8% of answers but recommended in 0.2%: about one recommendation for every 65 mentions. A quarter of brands (15 of 60) weren't recommended once in the month. The definition is strict on purpose, which is why the comparison with competitors matters more than the absolute number.

In the competitor table, the recommended rate sits next to visibility. In the demo workspace, the sportswear brand ranks 2 of 7 at 45% visibility over the past week and is recommended in 2.1% of checked answers. The leader sits at 51% visibility and is recommended in 5.4%. The gap in being picked is wider than the gap in being named. That's the one to work on.

Enlarge

Recommended rate and visibility side by side: second of seven at 45%, recommended in 2.1% of answers against the leader's 5.4%. Demo workspace; brand names blurred.

When it's low but Visibility is high: you make the shortlist but rarely win it. State plainly who you're the best choice for, and why, on pages AI can read. Back it with proof: pricing, outcomes and reviews.

3. Citation Share

Citation Share is the share of all citations in AI answers that point to your own website. Where Visibility asks "did it name me?", Citation Share asks "did it learn from me?"

How it's calculated:

Citation Share = (citations to your owned pages ÷ total citations) × 100

For any period, add up owned citations and all citations over the period, then divide. Don't average daily values; that gives a different number.

What good looks like: when a model answers, it runs searches behind the scenes and visits several pages. A Citation Share of 1.3% means about 1 in 100 of those sources is yours. Higher is better, up to a point: a 100% owned profile would be a warning sign, because no independent source vouches for you.

What normal looks like: single digits. Across 100 brands and 5.9 million citations, the median brand's own domains made up 6.8% of the sources AI cited (interquartile range 3.4% to 10.1%). Only 3 of 100 were above 20%. A citing answer lists a median of 8.6 sources, so being one source among many is the norm.

The demo workspace shows why Citation Share and Visibility are different signals. The sportswear brand is named in 44.5% of answers, but it owns only 2,254 of 88,785 citations in a week, a 2.5% share. Its most-cited own pages are its homepages, a campaign page, a kids' size chart and a buying guide. Help content earns citations.

Enlarge

The Citations view for the past week: a 2.5% citation share, with Perplexity as the model that cites the brand most. Demo workspace; brand names blurred.

Why it matters early: Citation Share is a leading indicator. It often rises before Visibility does. If your pages start getting cited this month, mentions tend to follow.

When it drops: your pages either don't exist for the prompts that matter, or AI can't extract them. See #13 and #14.

4. Sentiment

Sentiment is how positively AI describes your brand when it mentions you. Being mentioned a lot only helps if the framing helps you.

How it's calculated: every answer that mentions your brand is scored at two levels: the overall tone of the answer, and each sentence that names you. The headline score is a confidence-weighted balance of positive against negative phrases. Neutral phrases are counted but don't move it. Answers that don't mention you never dilute the score.

Sentiment score = positive weight ÷ (positive weight + negative weight) × 100

What good looks like: read the distribution, not just the headline. Half positive and half negative means AI is actively arguing about you. Uniformly neutral means it has nothing strong to say.

When it drops: open Sentiment themes. They surface the recurring claim or objection, such as an outdated price, a support complaint or a missing feature. Then fix it at the source: your own pages, or the review sites AI cites.

The demo workspace shows why you read the phrases, not just the score. The sportswear brand sits at 84% positive, down 3 points on the previous period. The recurring negatives are about price and fit: "you partly pay for the brand" and "fit varies". That isn't a reputation problem. It's two specific claims you can address with clear fit notes and value arguments on the pages AI cites.

Enlarge

Sentiment for the past week: 84% positive phrases. The negatives about price and fit tell you what to fix. Demo workspace; brand names blurred.

Example: a bank shows 70% positive overall. The negative share sits almost entirely on one recurring complaint about one fee. That's a narrow, fixable issue, not a reputation problem.

5. Position

Position is where your brand appears inside a single AI answer, compared to your competitors. Being tenth in a "top 10" list is very different from being the first recommendation.

How it's calculated: 3RD reads each answer for a numbered list. Your line number in that list is your position. If there's no list, it uses the order of first mention. Lower is better, and position 1 is the lead recommendation. Average position is the mean across the answers where you appear. Only tracked competitors count in the ordering.

How to read it:

When it slips: make your differentiators explicit for each prompt's intent, and earn a place on the comparison and "best of" pages that models take their ordering from.

6. Rank

Rank is your place among all tracked brands, sorted by Visibility, shown as "X of Y". It's the fastest answer to "are we winning or losing?"

How it's calculated: 3RD sorts your brand and your competitors by visibility for the current filter. Your place in that order is your rank. It's recalculated per period, so you can track movement. The dashboard shows it on the Competitor rank card and the Leaderboard.

Position vs. Rank: these get mixed up more than any other pair.

The trap: Rank is only as honest as your competitor set. Track too few brands and "1 of 3" becomes a comfortable illusion. Include the brands AI recommends instead of you, even if they aren't on your board's competitor slide.

Example: an insurance brand ranks 3 of 5. It's ahead of two smaller players and behind the top two. "We have a visibility problem" becomes "we have a specific gap to two specific competitors."

How prompts are weighted

Not every question is worth the same. These two metrics tell you which numbers to care about first.

7. Prompt tiers and targets

Prompt tiers sort every tracked question by commercial value: Must-Win, Core and Foundational, each with its own visibility target.

The tiers mirror the buying journey. A strong total driven by Foundational prompts while Must-Win lags is a warning, not a win.

What normal looks like: brands tend to score highest on the prompts they mark as most important. Across 97–99 brands, median visibility was 32.2% on Must-Win prompts, 24.2% on Core and 12.3% on Foundational. Must-Win beat Foundational for 83 of 96 brands. Part of that is how teams define the tiers: Must-Win prompts often sit closer to the brand's own category terms. If your Must-Win score is below your Foundational score, you're the exception, and it's worth a look.

Example: a bank scores well on the Foundational prompt "what is a savings account" and poorly on the Must-Win prompt "best savings account Denmark". The second gap is the one that costs money.

Est. Volume helps you rank prompts inside a tier. It's a demand score from 0 to 5, built from commercial intent, audience breadth and question clarity, and assessed in the prompt's own language.

8. Prompt coverage

Prompt coverage is the share of your tracked key prompts where your brand is mentioned at all. The dashboard shows it on the Key prompts coverage card, with the prompts you win and the ones you're absent from.

Why it matters: an average hides where you're missing. A software company can have strong coverage overall and still be absent from its three highest-intent prompts. The list of missing prompts is more useful than the percentage.

When it's low: start with Must-Win prompts where you're absent. They become your content and outreach list (#13 and #10).

Where the answers come from

These three tell you who is shaping the answers about you, and so which team can change them.

9. Owned vs. earned citations

Every citation is either owned (your website or a subdomain) or earned (someone else's site), and the split tells you which team fixes a gap.

The Attribution by provider card shows your owned share per AI assistant. Models often build their answers from different parts of the web.

Example: a furniture brand finds that 60% of its ChatGPT citations point to its own product pages. On Perplexity the figure is 20%, and the rest go to retailers and review sites. The two models need different work.

A low owned share means AI isn't learning from you. A high earned share on Must-Win prompts means a few third-party pages decide your answers. Healthy visibility needs both.

What normal looks like: in our data, the split differs by assistant. Google's surfaces cite brands' own pages most (median owned share 12.0% in AI Mode and 9.7% in AI Overviews), and the chat assistants least (5.8% in ChatGPT, 3.2% in Claude). One more thing worth checking: for 41 of 100 brands, AI cited competitors' websites more often than the brand's own.

10. Citation sources

Citation sources are the specific pages AI used to build its answers, each with a citation count, its share of all citations and the prompts it influences. They sit in the Citations view as Top cited sources.

A citation is a page the model used as a source, whether or not it named your brand. You can be mentioned without being cited, and cited without being mentioned.

Why it matters: AI models lean on a small set of trusted sources per category. One page cited across dozens of your Must-Win prompts shapes every answer built from it.

The same concentration shows up on your own site. For the median brand in our data, the 10 most-cited pages took 57% of all citations of that site, out of 193 different pages cited at least once (n = 98). Know your top 10, and keep them accurate and current.

Outreach opportunities come from this data: third-party pages that AI cites often for your prompts, that feature competitors, and that don't mention you yet. 3RD ranks them by how often AI cites them, which competitors they feature, how many of your prompts they influence and how valuable those prompts are. Each one comes with an outreach strategy for that page.

Example: a travel brand finds that one "best hotels in Copenhagen" list gets cited again and again on the exact questions it cares about, and the brand isn't in it. "We should do outreach" becomes one named target.

11. Topic visibility

Topic visibility is your Visibility for each group of related prompts, so you can see which subject areas you own and which you don't. The dashboard shows it as the Topics breakdown and on the Topics view of the Leaderboard, with change against the previous period.

Example: a snack brand is strong on recipe ideas and weak on healthy snacking, though both get similar demand. That's a specific content gap, not a general one.

Click into a weak topic before acting. Sometimes one market drags the whole number down.

Enlarge

Buyer questions grouped into topics and tracked per market. Basketball Shoes sits at 85.2% visibility, Socks at 10.4%. Demo workspace; brand names blurred.

Why AI picks a competitor

The metrics above tell you what's happening. These three tell you why.

12. Drivers strength (the opportunity radar)

Drivers strength scores the five things AI uses to choose a brand: content coverage, content quality, agent readiness (formerly crawlability), authority and outreach. Each gets a strength score from 0 to 100 per market, shown side by side so strengths and weaknesses are easy to compare.

Medians across 56–60 brands, scores as of October 19, 2026. Unmeasured drivers excluded.

What normal looks like: technical access is rarely the bottleneck. The weakest driver was authority or content quality for 42 of 54 brands, and agent readiness for none. Authority also has the widest spread (interquartile range 21 to 64), so it's what separates brands from each other.

The demo workspace follows the same pattern. In Denmark, the sportswear brand scores content coverage 47, content quality 45, agent readiness 76, authority 61 and outreach 50. Content quality and coverage are the weakest drivers, so the next fix sits with the content team, not with the developers.

Enlarge

Drivers strength for Denmark. Content coverage (47) and content quality (45) are the weakest, so content owns the next fix. Demo workspace; brand names blurred.

Why it matters: each driver points to its own kind of work, so whoever owns that work knows what's theirs. Diagnostics writes up the findings in plain language. Each insight connects the metric that moved, the likely cause and a recommended fix, with the AI answers behind it as evidence.

Example: a retailer scores well on content quality and poorly on crawlability. Technical barriers stop AI from reading pages that would otherwise get cited. That fix is cheap compared with the visibility it unlocks.

13. Content coverage and impact

Content coverage shows, for each prompt, how well your existing pages answer it, with the number of matching pages and an impact score for closing the gap.

• Coverage: how well your current content covers the prompt.
• Page matches: how many of your pages match the prompt.
• Impact: an opportunity score built from the prompt's tier, your current coverage and the prompt's volume.

Where to start: filter to Must-Win prompts with no matching page. Those are questions with commercial value that you have no answer for, and the highest-leverage content you can publish.

One pattern is worth knowing. High coverage with low Citation Share on a prompt means you have a page, but it isn't the source AI uses. That's a quality problem (#14), not a missing page.

Example: an electronics brand sees a high-impact prompt with zero matching pages on its own site. Nobody has to guess what to write next.

14. Content quality and agent readiness

Content quality scores whether AI can extract and quote your pages. Agent readiness tests whether AI can reach, recognise and read your site in the first place.

Content quality checks each page for five signals that make citation more likely:

Agent readiness (formerly called Crawlability) tests every site you track in four stages:

Each stage reads Good, Minor issues, Needs work or Critical. Each issue shows its impact, which sites and page templates it affects, and how to fix it.

Example: a B2B software brand's most thorough page on its topic also has the worst technical health in its library. That's a plausible reason a page that should win citations doesn't.

What to do next

15. Playbook

A Playbook is a goal-based plan: you set a target for a group of prompts, and 3RD generates the prioritized tasks to reach it. It turns the diagnosis into a to-do list instead of leaving the numbers to speak for themselves.

• The goal: a metric to raise for a defined set of prompts, for example "raise visibility to 70% on Employer Branding in ChatGPT."
• The tasks: generated from Diagnostics, agent readiness tests, content analysis and outreach opportunities, each with an estimated impact. You can add your own.
• The work: tasks run on the Task Management board (To do, In progress, Done), assigned to the people who own them.
• Progress: a trend chart on the goal shows whether the work is moving the number.

You can also ask 3RD Agent what to focus on given your goal. It tells you what needs doing, and on request it drafts the work: a page, a brief, an outreach email. You check it and publish. 3RD never publishes for you.

Example: a bank with a large gap between its Foundational and Must-Win scores gets a task to publish a comparison page for one high-value Must-Win prompt, with a brief and the search phrases models use. The diagnosis becomes a next action.

When a task is done, 3RD shows its measured effect: what changed on the prompts and page the task targeted, compared with the brand's other prompts over the same window. Take an example from the demo workspace. The sportswear brand completes a content quality task: a rewrite of one Danish outlet page ("Optimize lob sko outlet"). Afterward, the page score rises from 46 to 53. But the share of AI models citing the page drops from 18% to 0%. Visibility on its prompts stays at 100%, while the brand's other prompts move from 57% to 55%. The prompts moved in step with the rest, so 3RD reads it as drift rather than effect. It labels the result as measured values, not proof that the task caused the change.

Enlarge

The measured effect of a page rewrite, read against the brand's other prompts. 3RD calls it drift rather than effect. Demo workspace; brand names blurred.

What normal looks like: a mixed picture, which is why the control matters. Of 117 completed tasks with a control group (44 brands), 29% moved up by more than 5 points relative to the control, 26% moved down by more than 5, and 44% stayed within ±5. Read a task's effect against the control and over time, never from one before-and-after reading. To check whether a finished task moved the number, see how to prove a GEO change actually moved your AI visibility.

Supporting terms

These come up throughout the dashboard.

Which metrics should a CMO report?

A CMO doesn't need all 15. Report five, in this order, and keep the rest for the team.

1. Visibility on Must-Win prompts, against your nearest competitor. Are we on the AI shortlist when buyers decide? Report the Must-Win number, not the total.
2. Recommended rate, against the same competitors. When we're on the shortlist, does AI tell buyers to pick us?
3. Citation Share. The early signal: it tends to move before Visibility does.
4. Sentiment, with the top theme. Is AI describing us accurately? Report the one recurring theme, not just the score.
5. Rank, by market. Where we stand, market by market.

Then add the commercial line: AI-referred visits, how they converted, and pipeline that started there. Keep it separate from the five above, and label it as a floor. How to build that link is covered in how to connect AI recommendations and citations to revenue.

The first step is to connect your analytics. In the demo workspace, Google Analytics isn't connected yet, so the traffic view shows a connect screen. Once you connect your GA4 property, sessions from AI assistants appear in the same dashboard as visibility. Those clicks come from AI citing your pages, and they won't move in step with your visibility score. Report them side by side, not blended.

Enlarge

To see AI-referred traffic, you first connect your GA4 property. Here, analytics is not connected yet. Demo workspace; brand names blurred.

How often

Don't report daily numbers upward. AI answers vary from run to run, and a single day is noise. In our data, 7 in 10 brands had at least one day more than 5 points away from their 30-day average.

A one-page report you can copy

AI visibility: <month>, <market>

1. Must-Win visibility: <X>% (last month <Y>%) · nearest competitor <Z>%
2. Recommended rate: <X>% · <competitor> <Y>%
3. Citation Share: <X>% (last month <Y>%)
4. Sentiment: <X> · top theme: "<theme>"
5. Rank: <X> of <Y> in <market>

What we did: <3–5 shipped tasks>
What moved: <one sentence, with the prompts it affected>
What's next: <next 3 tasks and the goal they serve>

Commercial floor (measured separately): <AI-referred visits> · <conversions> · <pipeline started from AI visits>

In 3RD, Reports Studio builds and schedules this kind of report, and 3RD Agent can build one when you ask it to.

What not to report

• Total visibility on its own. It mixes Must-Win with Foundational and can rise while the questions that matter fall.
• A blended "AI ROI" score. It hides which part is measured and which part is estimated.
• Averages of daily rates. Sentiment, Citation Share and Recommended rate must be added up over the period, then divided.

Common mistakes when reading these metrics

• Reading the total instead of the tier. Check Must-Win before you celebrate.
• Confusing Position and Rank. Position is one answer. Rank is everything you track.
• A flattering competitor set. Rank and Position only count the brands you track.
• Including branded prompts. They inflate Visibility and Recommended rate, which is why 3RD leaves them out by default.
• Chasing 100%. On competitive prompts, beat your nearest competitor and keep the trend rising.
• Acting on a number without reading the answers. The metrics compress reality. The answers in Conversations are reality.

About the data

The "what normal looks like" figures come from customer brands tracked in 3RD between September 19 and October 18, 2026. They're anonymized and aggregated, and no brand, domain or prompt is named. Variance figures use 111 brands with a stable daily series and at least 1% mean visibility (732,000 answers). Citation figures use 100 brands (5.9 million citations). Recommendation and driver figures use 60 brands, at most three per customer organization. Task outcomes use 117 completed tasks with a control group across 44 brands. The sample leans toward the Nordics, and some brands are tracked by agencies. These are observational data, not experiments.

FAQ

What is the most important AI visibility metric? Visibility on your Must-Win prompts. It tells you whether AI puts you on the shortlist when buyers are close to a decision. Read it together with Recommended rate, which tells you whether you win that shortlist.

What is a good AI visibility score? 3RD's recommended targets are 75% on Must-Win prompts, 50% on Core and 35% on Foundational. On competitive prompts, a rising score at or above your nearest competitor matters more than any absolute number. For reference, the median Must-Win visibility across 97–99 brands in 3RD was 32.2%.

What's the difference between Visibility and Citation Share? Visibility counts answers that name you. Citation Share counts how many of the sources behind the answers are your own pages. You can be named without being cited, and cited without being named.

What's the difference between Position and Rank? Position is where you appear inside one AI answer, where 1 is first. Rank is your standing among all tracked brands by visibility, shown as "X of Y".

Can these metrics prove AI visibility drives revenue? Not on their own. They show whether AI names, recommends and cites you. Revenue needs a separate link through AI-referred traffic and pipeline, and part of AI's influence never produces a click. Our guide to connecting AI recommendations to revenue shows what you can measure and what you can't.

Want to see these numbers for your brand? Start a 7-day free trial, no card needed, or read how 3RD measures AI answers.

Written by

Magnus Windeløv

Head of Marketing

Magnus leads marketing at 3RD. He has spent his career in demand generation, growth and SEO for B2B software companies and consumer brands.

View profileLinkedIn

Reviewed by

Nikolaj Peters

Solution Architect

Nikolaj is a senior SEO specialist and Solution Architect at 3RD. He has more than ten years in SEO and digital performance, including years leading the owned media and SEO team at iProspect.

View profileLinkedIn