I've had a version of this exchange with more marketing leaders than I'd like to admit. One flattering screenshot is not measurement, it's a horoscope. AI answers vary by phrasing, session, engine, and model version, so a single check tells you almost nothing.

After eight-plus years building measurement systems for search, I've settled on seven metrics that turn AI brand visibility from a vibe into a number you can manage, report, and improve. This guide explains each one in plain language, how to track it, and what "good" looks like.

Why Measurement Comes Before Optimization

Every AI visibility failure I audit shares a root cause: the brand optimized blind. They rewrote pages, chased schema, hired agencies, without a baseline, so they could never prove what worked. Measurement isn't the boring prerequisite; it's the strategy. When you can see which questions you lose and why, the fixes become obvious.

The 7 Metrics That Matter

1. Share of Model (SoM)

What it is: Of all relevant AI answers in your category, the percentage that mention your brand versus competitors. It's market share for the answer economy.

How to track: Define 30–50 realistic buyer questions. Query the major engines repeatedly (tools automate this). SoM = your mentions ÷ total brand mentions.

What good looks like: Category leaders typically hold outsized SoM, AI engines concentrate recommendations more than search results pages ever did. If you're under 10% in your own niche, treat it as a fire alarm.

2. Mention Position

What it is: When an AI lists options, where you appear. First-mentioned brands anchor the user's shortlist; fifth mentions are trivia.

How to track: Log average position across your question set, per engine.

What good looks like: Consistent top-two placement for your core specialty questions.

3. Sentiment Score

What it is: The tone of your mentions. "The top-rated choice for small teams" and "a cheaper option with mixed support reviews" are both mentions; only one sells.

How to track: Dedicated platforms score sentiment automatically and, crucially, trace it to sources, review clusters, forum threads, articles. That traceability is what makes sentiment fixable. Polyvalent's approach here is a good reference point, and their comparison of the best AI visibility tools shows how sentiment depth varies dramatically between dedicated platforms and legacy SEO suites.

What good looks like: Predominantly positive descriptors with specific strengths named. Watch for hedging words: "however," "some users report," "mixed."

4. Citation Rate

What it is: How often engines that show sources (Perplexity, AI Overviews, ChatGPT with browsing) cite your domain as evidence.

How to track: Monitor citations for your target questions monthly. This metric responds fastest to on-page fixes, direct answers up top, FAQ schema, fresh dates, making it your best early-progress indicator.

What good looks like: Rising citations on your money questions within 4–8 weeks of content improvements.

5. Entity Accuracy

What it is: Whether AI engines describe your business correctly, right services, right locations, right positioning. Hallucinated pricing or phantom services quietly kill conversions.

How to track: Monthly, ask each engine "What is [brand] and what does it offer?" Log errors. Fix the sources: inconsistent directory listings, outdated pages, thin About content.

What good looks like: Zero material errors across engines.

6. Competitor Gap Index

What it is: The list of questions where competitors appear and you don't. Less a number, more a battle map.

How to track: From your question set, extract every answer mentioning a rival without you. Group by theme. Each theme is a content or reputation project.

What good looks like: The list shrinking quarter over quarter.

7. Downstream Lift

What it is: The business echo of AI visibility: branded search volume, direct traffic, and "how did you hear about us?" responses citing AI assistants.

How to track: Add "AI assistant (ChatGPT, etc.)" to your lead-source dropdown today, it costs nothing and produces the ROI evidence every budget conversation needs. Watch branded search trends alongside your SoM curve.

What good looks like: Branded demand rising in step with Share of Model.

Building Your Monthly Scorecard

Keep it to one page:

• Headline: Share of Model (trend arrow vs last month)

• Row 2: Sentiment score + top negative source flagged

• Row 3: Citation rate on top 10 money questions

• Row 4: Competitor gap count (and the one gap you'll close this month)

• Row 5: Downstream lift (branded search + AI-attributed leads)

Five rows. Any executive can read it in ninety seconds, and it tells a complete story: presence, perception, proof.

FAQs

1. How many questions should my tracking set include? Start with 30: ten broad category questions, ten specific use-case questions, ten comparison questions. Expand once the workflow is stable.

2. How often should I measure? Monthly for the full scorecard; weekly spot-checks on citation rate during active optimization sprints.

3. Can I do all this without paid tools? The manual version works for baselines, but sampling variation makes trends unreliable. Tools earn their cost through repeated, consistent querying and sentiment tracing.

4. Which metric should a beginner start with? Share of Model. It's one number, it's motivating, and every other metric explains its movement.

5. My SoM dropped suddenly with no changes on our side. Why? Likely a model update or a competitor's reputation surge. This is exactly why monthly tracking matters, sudden drops caught early are recoverable; discovered late, they're entrenched.

Conclusion

AI visibility stops being mysterious the moment you measure it properly. Seven metrics, Share of Model, mention position, sentiment, citation rate, entity accuracy, competitor gaps, and downstream lift, turn an invisible battlefield into a one-page scorecard you can manage like any other channel. Build your question set this week, capture your baseline this month, and review the scorecard every thirty days. What gets measured gets improved, and in AI search, the improvers are quietly taking the answers everyone else assumes they own.