What is a good AI visibility score?

No study or standards body has published a target AI visibility score. A 0 to 100 scale invites letter grades, but the scale is set by each tool's designer. A 60 from one tool and a 60 from another are not the same result.

The reason is structural. Any single AI visibility score averages answers given with live web retrieval and answers given without it, and nobody outside the engine sees which was which. As of October 2026, ChatGPT decides per query whether to search, and Claude.ai search is a user toggle. So a score is a model of your visibility, not a direct measurement of it. Our guide on what AI visibility scores average walks through the components.

That leaves two comparisons that hold up:

• You against your own baseline, measured the same way each time.
• You against like-sized competitors, measured on the same prompts, engines and formula.

A good score is one that moves the right way on both.

Five inputs move a score even when nothing about your brand has changed:

• The prompt set. Buyer questions a tool writes for you differ from the ones you write. Engines also rewrite prompts into their own search queries, so any prompt list is a proxy.
• The engines. Google states that AI Overviews and AI Mode "may use different models and techniques" and can return different links. Treating them as one surface blends two products.
• Retrieval state. An answer from model memory reflects a training cutoff. An answer with live search reflects the index today.
• The formula. Weights for mentions, citations, accuracy and sentiment are product decisions, not standards.
• The run date. Vendors update models and defaults without notice.

This is why AI visibility tools disagree about the same brand. Neither tool has to be wrong for the numbers to differ.

Three findings make cross-industry averages unreliable.

Citations concentrate on big sources

A 2025 study of AI search systems (arXiv:2507.05301) found the top 20 news sources accounted for 67.3% of all citations for OpenAI models, using data from March to May 2025. An industry average built on raw citation counts mostly reports brand size. A mid-market brand will look weak against it whatever it does.

Engines behave very differently

In OtterlyAI's 2026 experiment, 2,718 of the 3,934 citations it tracked for one test site came from Microsoft Copilot, while ChatGPT, Gemini, AI Overviews and AI Mode almost never cited it. A blended score would have hidden that split entirely.

Prompt samples undercount

In the same test, Microsoft's own AI Performance report counted 42,817 citations for that site, against the 3,934 the prompt tracker logged. The windows may not match, but the lesson stands: a prompt-based score is a sample.

There is also no ground truth for rank. Google's AI report exposes impressions only, so there is no average position to benchmark against. For more, see whether industry benchmarks can be trusted.

Build your own reference point instead of borrowing one.

1. Freeze the method. Fix the prompts, engines, retrieval settings and scoring formula. Write them down.
2. Record a baseline. Run it more than once before you set any target, because single runs vary.
3. Pick three to five like-sized competitors. Match on market, category and brand size. The like-sized competitor playbook covers how to choose them.
4. Read each engine on its own. Per-engine results show where you are absent. See one blended score vs per-platform scores.
5. Track direction, not a threshold. Judge the trend over several runs.

Say you run a regional accounting firm. Benchmarking against a national brand tells you little. Benchmarking against three other regional firms on the same ten buyer questions tells you where you stand and which engine to look at first.

Expect competition to matter. C-SEO Bench found that as more sites adopted the same tactics, the gains shrank, which it described as zero-sum.

Before reacting to a change, rule out method changes.

• Formula revisions. 99Visibility revised its accuracy formula on 18 September 2026, and scores from before that date are not comparable with scores after it. Any vendor can do the same. See why old and new scores don't compare.
• More prompts. A well-built score should give the same result at the same error rate whatever the sample size. If yours fell when you added prompts, question the formula first.
• Partial runs. If an engine errors, results from a handful of answers can swing wildly.
• Model updates. A new default model can change answers overnight.

If none of these apply, work through why a score drops.

A score tells you where to look. The fixes with the strongest evidence are about access, and they are in your control:

• Check that robots.txt allows each retrieval bot by name: OAI-SearchBot, PerplexityBot and Claude-SearchBot. Blocking GPTBot affects training only.
• Confirm your firewall or CDN does not return 403 to crawlers that robots.txt allows.
• Make sure prices, product names and key facts appear in the HTML before JavaScript runs.
• State specific facts plainly on a fetchable page, so engines have a sentence they can quote.

None of this guarantees citations. It removes the blocks that make visibility impossible. For a quick starting snapshot, the free checker on this site shows whether AI engines name you for up to three buyer questions.

Is 50 out of 100 a good AI visibility score? It depends entirely on the tool, the prompts and who you compare against. A 50 could be strong in a category dominated by large brands and weak in a niche with few competitors. On its own, the number carries no verdict. Compare it with your previous runs and with like-sized competitors scored the same way, then read which engines drive it.

What is the average AI visibility score? No credible average exists. Published figures come from different tools using different prompts, engines and formulas, so they cannot be pooled. Citation data also concentrates on the largest sources, so any cross-market average mostly reflects brand size. Treat a quoted average as one vendor's sample, not a benchmark.

Should I compare my score with the market leader? Only as context, not as a target. Large brands dominate AI citations largely because they are large, and that advantage is not something a page change reverses. You learn more from three to five competitors of similar size in your market, measured on identical prompts and engines.

Can I set a score target with my agency? You can, but tie it to a fixed method and a direction, not a universal number. Agree the prompt set, engines and formula in writing, record a baseline, and judge movement over several runs. Avoid targets that imply a guaranteed outcome, because no one controls what the models say.