What Is an Omission Rate in AI Brand Monitoring, and How Is It Calculated?
An omission rate is the share of relevant AI answers that leave your brand out when it should reasonably appear. It is the flip side of an appearance rate, but the useful versions depend on a careful denominator. Count only questions where your business is a legitimate answer, for example where you serve that city, segment or need. For the strictest version, count only answers that actually recommend at least one brand, so that refusals and general advice do not inflate it. Two variants are especially useful. Peer omission counts answers that name your main competitors but not you. Fact omission counts answers about your brand that leave out a key fact, such as a free plan or a certification. Report each with its sample size and a likely range.
Why measure omission rather than appearance?
The numbers are complements, but omission focuses attention on where you are missing and why. It is also easier to act on. A list of omitted answers, grouped by topic, is a list of gaps. Appearance rates invite celebration; omission rates invite diagnosis. They also make one common measurement error visible: counting questions you should never win. A roofing company with no commercial work is not "omitted" from answers about commercial roofing.
How is each version calculated?
| Measure | Numerator | Denominator | What it tells you |
|---|---|---|---|
| Simple omission rate | Eligible answers that do not name you | All answers to eligible prompts | Overall absence where you fit |
| Conditional omission rate | Answers that name at least one brand but not you | Answers to eligible prompts that name at least one brand | Absence when the assistant does recommend someone |
| Peer omission rate | Answers naming at least two of your top three competitors but not you | Answers naming at least two of your top three competitors | Absence from your real competitive set |
| Fact omission rate | Answers about you that leave out a key fact | Answers that mention you | Whether important facts come through |
Eligibility is the crucial step. For each prompt, decide in advance whether you are a legitimate answer: right location, right segment, right product. Mark ineligible prompts and exclude them. Without this step, omission rates are inflated by questions you should not expect to win.
Recommendation answers only, in the conditional version, removes answers where the assistant declined, gave general advice or asked a follow-up question. Those are a different outcome, covered in what refusal means in AI search and how it differs from a no-recommendation.
What does a worked example look like?
Consider an illustrative software company tracking 60 prompts, run five times each, giving 300 answers.
- Eligibility: 45 prompts, 225 answers, fit the company. The other 15 concern segments it does not serve.
- Simple omission: the company appears in 80 of the 225 eligible answers, so it is omitted from 145. That is 64%, with a likely range of about 58% to 70%.
- Conditional omission: 200 of the 225 eligible answers recommend at least one brand. The company is missing from 120 of them, 60%, with a range of about 53% to 67%.
- Peer omission: 150 answers name at least two of the company's three main competitors. The company is missing from 100 of them, 67%, with a range of about 59% to 74%.
- Fact omission: of the 80 answers that mention the company, 60 leave out its free plan, 75%, with a range of about 65% to 83%.
The ranges are Wilson intervals, a standard way to put bounds on a proportion. Since repeated runs of the same prompt are related, treat them as a minimum. How to count mentions and put intervals on them is covered in how to count brand mentions in AI responses with keyword matching and confidence intervals.
The peer omission rate is the most telling here. When assistants name two of your three main rivals and not you, you are missing from a set you clearly belong to. Fact omission shows a second problem: even when the company is named, its free plan, a key buying reason, rarely comes through.
How do you use omissions to find gaps?
- List every omitted answer with its prompt, assistant and cited sources.
- Group by topic, question type and assistant. Omissions usually cluster, such as on comparisons, a specific use case or one assistant.
- Read the cited sources in omitted answers. If they are lists and reviews that do not include you, the gap is third-party. If they are competitor pages answering a question you do not answer, the gap is your own content.
- Check fact omissions against your pages. If the fact is missing or hard to find on your site, that is the first fix.
- Track omission by group over time, rather than one overall number.
What to check first when a competitor keeps appearing instead of you is covered in what to check first when AI recommends a competitor over your brand.
What mistakes should you avoid?
- No eligibility filter, which counts questions you should not win as omissions.
- Counting refusals as omissions in the conditional rate.
- Missing name variants, which counts answers that name you under a variant spelling as omissions.
- Mixing assistants into one rate when their list lengths differ.
- Single runs, which make omission look random from week to week.
How does omission fit with other brand metrics?
Omission is one of four measures that together describe how assistants treat a brand. Appearance and omission say whether you are named. Stance says how you are described when named: recommended, listed, compared unfavourably or warned against, as covered in whether a negative mention counts as AI visibility. Accuracy says whether the facts about you are right. Fact omission sits between the last two, because a missing fact is not wrong but can still cost the sale. Report all four side by side. A falling omission rate paired with a rising share of negative mentions is not progress.
How often should omission be measured?
With the rest of your tracking, on a fixed prompt set, monthly for most brands. Report the peer omission rate alongside the overall rate. It can move before the overall number does, because it responds to changes in the specific lists and sources your competitors appear in.
What might an illustrative case look like?
Consider an illustrative accounting firm whose tool reported "42% AI visibility" across 50 prompts, which the team read as being missing from 58% of answers. Recalculated with eligibility, 18 of the 50 prompts concerned services it did not offer, where it never appeared; on the 32 eligible prompts its simple omission rate was about 34%. The real gap showed in its peer omission rate: in answers naming at least two of its three main competitors, it was missing 80% of the time. Grouping showed the omissions clustered on "small business tax accountant in [city]" questions, where answers cited a local directory and a "best accountants" list that did not include the firm. The firm completed its directory profile and approached the list's publisher. Peer omission on those questions fell to 55% over the next quarter.
What do people ask about omission rates?
Is omission rate just one minus the appearance rate?
The simple version is, when both use the same eligible denominator. The conditional, peer and fact versions answer different questions.
What is a good omission rate?
It depends on the category and list lengths. Compare yourself with competitors on the same prompts, and track your own trend.
Should omission include answers that mention us only in passing?
No. A mention counts as an appearance. Measure how you are described separately.
Published by The Kovmere Team · updated Sep 29, 2026 · More resources