# Benchmarking: What Is a Good AI Visibility Score?

> What counts as a good AI visibility score? Why no universal benchmark exists yet, and how to build fair internal and competitor benchmarks.

Source: https://mentionx.ai/learn/measurement/benchmarking-good-ai-visibility-score  
Author: Dipuck Jones, founder of MentionX  
Updated: 2 November 2026  
Level: Senior Leader

**There's no universal "good" AI visibility score, because every tool uses a different formula, prompt set and engine mix, so a 42 in one tool can't be compared with a 42 in another. Our view: judge your score against three benchmarks measured the same way: your own baseline, your direct competitors, and the category leader on the same prompts.**

*Points marked "Our view" are interpretation; examples are illustrative. We're building category benchmarks from MentionX scan data and will add them here when the sample is large enough.*

> **Key takeaways**
>
> - AI visibility scores aren't comparable across tools, because prompts, engines and formulas differ.
> - Benchmark against yourself over time, against named competitors on the same prompts and against your own segments.
> - A good score is one that beats your baseline and your closest competitors on the prompts that matter to buyers.
> - Be sceptical of published industry averages unless the method is published with them.

*Checked on 31 October 2026. "Our view" marks MentionX opinion. Ledgerline is a fictional company used for illustration.*

## Why can't you compare scores across tools?

An AI visibility score combines inputs such as mention rate, citation rate, recommendation rate, position and sentiment, with weights each tool chooses; see [how a score is calculated](https://mentionx.ai/learn/measurement/ai-visibility-score-calculated). Change the prompts, engines, runs or weights, and the number changes. *Our view:* a score is only meaningful next to its method.

## What makes a fair benchmark?

| Condition | Why it matters |
|---|---|
| Same prompt set | Different prompts produce different visibility |
| Same engines and modes | Engines cite and recommend differently |
| Same dates and locations | Answers change over time and by region |
| Same number of runs | Single runs swing |
| Same formula | Weights change the number |

## Which three benchmarks should you use?

1. **Your baseline.** Your score in the first full month of tracking. Progress is measured against it.
2. **Direct competitors.** Three to six rivals on the same prompts; see [tracking competitors](https://mentionx.ai/learn/measurement/tracking-competitors-ai-search).
3. **Category leader.** The brand most often recommended on your prompt set. *Our view:* this shows the realistic ceiling for your category today.

## What does "good" look like in practice?

*Our view:* a good score is one that's above your direct competitors on the prompt groups that matter most to revenue, and rising against your baseline. Being the leader on a few high-intent prompt groups can be worth more than a high average.

| Situation (fictional) | Our view |
|---|---|
| Score below most competitors, rising | Good trajectory; keep going |
| Score above competitors, flat | Defend; watch new entrants |
| High average, low on buying prompts | Rebalance effort to high-intent prompts |
| Big swing in one month | Check sample size before reacting |

## How do you set targets?

*Our view:* set targets on component rates rather than the composite score, because they're easier to act on. For example: "raise recommendation rate on comparison prompts in ChatGPT from our baseline by a set amount within two quarters." Choose the amount after you've seen two or three months of normal variation; see [statistical confidence](https://mentionx.ai/learn/measurement/statistical-confidence-ai-visibility-tracking).

## What about published "industry averages"?

*Our view:* be sceptical. Ask what prompts, engines, dates, runs and formula sit behind any average, and whether the sample includes brands like yours. Without those details, an average can't tell you whether your score is good.

## What will MentionX benchmarks include?

When published, we intend to show, by category: the distribution of scores, component rates, sample size, prompt set description and dates. Until then, this page describes method only and makes no claims about typical scores.

## Which pitfalls should you watch for?

- Comparing scores from different tools.
- Reporting a score without its components.
- Benchmarking against brands in a different category.
- Changing the prompt set and comparing to the old baseline.

## How do you present benchmarks to leadership?

*Our view:* show a simple table with your score and each competitor's, each with component rates and the sample size, plus your baseline. Add one line on what the gap to the leader is made of, such as "we're named as often but recommended half as much". That points to action instead of debate about the number.

## How often should you rebenchmark?

*Our view:* monthly for tracking, with a fuller review each quarter where you check the competitor list, prompt set and weights. If you change any of them, reset the baseline and keep the old series for reference.

## How do prompt groups change the picture?

*Our view:* averages hide strengths and weaknesses. Break your benchmark down by prompt group, such as "best" prompts, problem prompts, comparisons and alternatives, and by engine. You may lead on how-to prompts yet trail on buying prompts. Those breakdowns are where targets and actions should come from, because they map to specific pages, sources and positioning choices.

## How should agencies benchmark clients?

*Our view:* agree the method with the client before reporting, and keep it fixed. Avoid comparing clients against each other's scores unless they share a category and prompt set. Report progress against each client's own baseline and direct competitors, which is fairer and easier to explain.

## What questions should you ask a vendor about their score?

*Our view:* before trusting any score, ask: Which prompts are included, and who chose them? Which engines, modes and locations? How many runs per prompt? What are the weights, and can I see the components? Can I export the raw answers? How are brand name variants matched? A vendor that can answer all of these gives you a score you can benchmark. One that can't is giving you a number without a method, and you shouldn't set targets or budgets on it. The same questions apply to MentionX scores, which is why we document our method on [the score page](https://mentionx.ai/learn/measurement/ai-visibility-score-calculated).

## Exercise

Calculate your score and two competitors' scores on the same 30 prompts, three runs each, in one engine, using one formula. Write down the component rates and the gap to the leader.

## Checklist

- Same prompts, engines, dates, runs and formula
- Baseline recorded
- Competitor benchmark monthly
- Targets on component rates
- Sample sizes shown
- No cross-tool comparisons

## From MentionX data: why one number misleads

**Method.** Read-only aggregate queries over MentionX scans from 22 August to 29 October 2026 (UTC): 2,217 engine runs (2,154 completed) across 18 anonymised projects and six engines.

Scores blend engines and intents that behave very differently. In our sample, mention rates by engine were:

| Engine | Brand mentioned | Answers with no citations |
| --- | --- | --- |
| Gemini | 47.3% | 50.7% |
| OpenAI | 42.9% | 30.2% |
| Perplexity | 31.5% | 4.1% |
| AI Mode | 31.3% | 12.6% |
| Copilot | 26.0% | 47.0% |
| DeepSeek | 24.0% | 99.0% |

And by prompt intent:

| Intent | Brand mentioned | Share of voice |
| --- | --- | --- |
| Brand-specific | 100% | 64.3 |
| Comparison | 91.9% | 44.6 |
| Recommendation | 29.2% | 26.8 |
| Best-of | 26.6% | 22.7 |
| Problem | 5.7% | 7.7 |

*Our view:* a single score that blends comparison prompts, where brands are named almost every time, with problem prompts, where they rarely are, says more about the prompt mix than about the brand. Benchmark by intent and engine first, then summarise.

**Limits.** A convenience sample of mostly B2B software projects, run through engine APIs, which can behave differently from the consumer apps. Perplexity, Copilot and DeepSeek figures come from a single day (26 October 2026), so treat them as a snapshot, not a trend.

## Worked example: Ledgerline sets its first benchmark

*Ledgerline is a fictional accounts-payable tool. Figures are illustrative.*

Ledgerline tracks 80 prompts in four engines. Its blended score looks low, and leadership asks whether that's bad.

The team splits the data. On comparison prompts, Ledgerline is named about as often as Competitor A and Competitor B. On best-of prompts, it trails Competitor A in two engines. On problem prompts, no brand is named often. Ledgerline sets three benchmarks: its own baseline by engine and intent, the gap to Competitor A on best-of prompts, and its mention rate on a segment of prompts for mid-sized finance teams. Targets are set as changes against each benchmark, not as a single score.

> **Pro move:** Freeze the prompt set for at least a quarter. Changing prompts mid-way makes old and new scores incomparable.

## How do you build a baseline?

Pick a prompt set, engines and run count, then measure for at least two or three rounds before setting targets. AI answers vary from run to run, so a single round can mislead. Record the range for each metric, not just the average. The baseline is the range you expect when nothing changes.

## How do you benchmark against competitors fairly?

Use the same prompts, engines, dates and runs for every brand. Don't use prompts that name your brand. Choose three to five rivals your buyers actually consider. Report each brand's mention rate and citation share by engine and intent. Gaps between you and a rival on specific prompt groups are more useful than overall rankings.

## What about benchmarks by segment?

Split prompts by buyer segment, product line or region if they matter to your business. A brand can lead in one segment and be absent in another. Segment benchmarks also make targets more realistic, because they focus on the prompts your team can influence.

## How do you explain variance to leadership?

Show ranges and explain that engines produce different answers to the same prompt. Point out that some engines rarely cite sources, so citation metrics there are unstable. Highlight changes only when they persist across rounds. *Our view:* leadership trust grows when you show the uncertainty honestly, rather than presenting every small movement as a win or a loss.

## When should benchmarks change?

Rebuild benchmarks when you add engines, change prompts substantially or enter new markets. Keep the old series for reference and start a new baseline. Note the change in reports so trends aren't misread.

## How do benchmarks feed planning?

Use the gaps to choose work. If you trail a rival on best-of prompts in one engine, look at the sources cited there. If you're absent from problem prompts, consider content that answers those problems directly. Set a target for each gap and review it at the next benchmark.

## How should you document the method?

Write down the prompt set, engines, run count, dates, scoring formula and any exclusions. Store it with the results. Anyone reading a benchmark months later should be able to see exactly how it was produced, and repeat it if needed.

## How do you handle new engines?

When a new engine becomes important to your buyers, add it to tracking as a separate series. Don't fold it into your existing blended score straight away, because it will shift the number for reasons unrelated to your work. Build a baseline for the new engine over a few rounds, then decide whether to include it in summaries.

## What should a benchmark report contain?

Keep it to one page. Show your metrics by engine and intent against the baseline range, the gap to each tracked competitor, and the two or three changes that persisted since the last report. Add a short note on the method and any changes to it. End with the actions planned for the next quarter.

## Common questions

**What is a good AI visibility score?** One that beats your own baseline and close competitors on the prompts that matter.

**Can I use another company's average as a target?** Only if their method and prompts match yours, which is rare.

**How often should I rebenchmark?** Quarterly for most teams, or when the prompt set changes.

## Related

- [AI visibility score: how it's calculated](https://mentionx.ai/learn/measurement/ai-visibility-score-calculated)
- [AI visibility KPIs](https://mentionx.ai/learn/measurement/ai-visibility-kpis-marketing-teams)

## Sources

This page describes a benchmarking method. No primary source defines a standard AI visibility score. For how Google's AI features select supporting links, see Google, [AI features and your website](https://developers.google.com/search/docs/appearance/ai-features).
All accessed 30 Oct 2026.

## Frequently asked questions

### Is 50 out of 100 a good AI visibility score?

It depends on the formula, prompts, engines and category. Our view: compare against competitors on the same prompt set, not against a universal number.

### Are there published industry benchmarks?

Not that we can verify for a common formula. Scores from different tools aren't comparable.

### What benchmark should I start with?

Your own baseline and your top three competitors, measured the same way, each month.

---

Written and reviewed by Dipuck Jones for MentionX Learn. The HTML version of this page: https://mentionx.ai/learn/measurement/benchmarking-good-ai-visibility-score