Own-brand presence in AI answers runs from 88.8% to 2.0% across seventeen industries, and industry itself explains almost none of it.
Own-brand presence in AI answers runs from 88.8% at the top of our industry table to 2.0% at the bottom, a forty-fold spread measured across 11,554 production runs on five engines. Industry itself explains very little of that. Four structural properties explain most of it.
Presence is the share of completed prompt runs in which the brand was mentioned, with each brand evaluated only on its own prompt set. We grouped own brands by industry and kept every industry carrying at least 100 own-brand evaluations. The table below is the production corpus only; staging was held back and used as a replication check on the shape of the findings, never pooled into the numbers.
One caution belongs before the table rather than after it. In this corpus one brand usually equals one industry row. These are case studies with volume, not population estimates of an industry. A row reading 61.2% describes one fine jewellery brand measured 129 times, not fine jewellery. We have published the industry labels and withheld the brand names, which are client information.
| Industry | Evaluations | Presence | Average position | Sentiment |
|---|---|---|---|---|
| Sports events | 169 | 88.8% | 1.33 | 78.9 |
| Retail electronics | 310 | 81.6% | 1.78 | 71.1 |
| Sports league | 407 | 79.4% | 1.47 | 75.0 |
| Fine jewellery | 129 | 61.2% | 1.62 | 81.4 |
| Food delivery | 695 | 57.8% | 2.02 | 72.5 |
| Monitored alarms | 1,352 | 57.6% | 1.28 | 77.1 |
| Dental, destination clinic | 120 | 55.0% | 1.44 | 79.1 |
| Renewable energy | 225 | 51.6% | 2.89 | 78.5 |
| Events and entertainment | 265 | 36.2% | 2.79 | 74.8 |
| E-commerce snacks | 1,357 | 30.4% | 2.15 | 70.9 |
| Sports technology | 176 | 27.3% | 2.79 | 72.3 |
| Consulting | 798 | 26.9% | 1.99 | 75.7 |
| DIY retail | 210 | 14.8% | 2.58 | 70.1 |
| Wellness tea | 248 | 13.7% | 1.62 | 76.1 |
| Dental, local practice | 275 | 6.9% | 2.11 | 73.0 |
| Legal | 1,357 | 5.0% | 5.01 | 74.6 |
| Outdoor cooking | 100 | 2.0% | 3.50 | 61.0 |
Sentiment is the least useful column on the page and we have printed it anyway. Fourteen of the seventeen rows sit between 70 and 79 on a 0 to 100 scale. Presence and position discriminate between brands; sentiment mostly does not.
Read down the table rather than across it and the same four things keep deciding the outcome.
Unique, ownable names occupy the top of the table. Generic or colliding names occupy the bottom. This is the clearest single pattern in the data, and it has a mechanism behind it: a model that cannot confidently establish who a name refers to declines to name it. One brand in this corpus, a logistics software company, carries a name shared with a far larger consumer financial brand. Its measured visibility is distorted in both directions at once — some of the mentions belong to the other company, and some of its real visibility is lost to the same confusion. We have withheld it from the table rather than publish a figure we cannot attribute cleanly.
Small fields with an obvious leader return 53% to 58% presence. Fragmented fields occupied by very large incumbents pin challengers between 5% and 30%, because the shortlist slots go to the incumbents. The renewable energy row at 51.6% is the pattern working in a brand's favour: a young, distinctively named business in an under-written category can own the answers cheaply.
Industries whose buying questions are asked specifically do far better than industries asked about generically. Legal's average position of 5.01 is what happens when even a mention lands deep inside an institutional list.
Every industry above 50% in this table is consumer-facing. The strongest business-to-business showing is consulting at 26.9%. That is the industry-level expression of a segment gap we measure directly elsewhere.
We should be plain about the status of these four. They are read off the table, not fitted to it. A regression across the 31 brands carrying 50 or more evaluations would test them properly. We have not run it.
5.0% presence at an average position of 5.01, across 1,357 evaluations, despite heavy prompt coverage. Answers to legal questions are dominated by institutional sources, professional bodies, directories, and the large full-service firms. Position matters more here than anywhere else in the table: on the rare occasions the brand is named, it is named fifth.
The strategic reading is uncomfortable but clear. There is no realistic campaign to win a head term like the generic search for a solicitor. There is a realistic campaign to be the named specialist on a narrow question, and that is a different content programme with a different budget.
Two rows in this table are the same profession. A destination clinic operating as a brand sits at 55.0%. A local practice sits at 6.9%. Same trade, same country, eight times apart. What separates them is not dentistry. It is review-platform depth and branding: the winner is surrounded by clinic-review citations that the local practice does not have.
This is the most important thing in the report. Every other row can be explained away as innate — some categories are simply better known than others, and nothing a brand does will change that. This pair cannot. Two businesses of broadly comparable size, in one profession, separated by an order of magnitude on a variable that is built rather than inherited. It is the strongest evidence in our corpus that the entity layer is buildable.
It is also one case, and we treat it as one case. We have a larger test planned against a multi-site clinic group, which is the honest way to find out whether this generalises.
Score yourself on the four properties before you commission any AI visibility work, because they will decide most of your result before the first prompt runs. Is your name unambiguous, or does it collide? Is your category concentrated around two or three giants, or is it open? Are your buyers asking specific questions or generic ones? Are you selling to a consumer or to a professional buyer?
If you score badly on all four, the useful work is not content optimisation. It is entity work: getting the third-party references, review depth and unambiguous naming that make you a resolvable thing before you try to be a recommended one. If you score well on all four and are still not appearing, something is wrong that is worth diagnosing.
The one thing this table should not be used for is benchmarking. Sports events and leagues at 79% to 89% are outliers built on unique entities, encyclopaedic coverage and heavy editorial attention. They make excellent demonstrations and terrible comparators. If a vendor shows you a figure in that range as evidence of what is achievable, ask what kind of entity produced it.
A regression across the 31 brands with 50 or more evaluations, testing the four properties against presence, would either support the reading or break it. It has not been run, and until it has, the four properties are a well-supported description rather than a model.
A second dental-style pair, in a different profession, would move the buildable-entity claim from one case to a pattern. A single pair that went the other way would weaken it considerably.
More brands per industry row would change what this table can be used for. At the moment it is a triage rubric. With five or six brands behind each row it would become a benchmark, and we would report it differently.
Corpus. 11,554 completed production prompt runs and 10,458 staging runs, between 15 April and 21 August 2026, across five answer engines: ChatGPT, Claude, Gemini, Perplexity and Google AI Overviews. Approximately 110 own brands, 72 of them in production, plus 153 tracked competitor brands, 895 prompts, 65 anonymous cold scans. 22,012 runs in total. Country mix is Ireland and Great Britain heavy at 86% of runs, with Hong Kong, South Africa and United States tails, so the findings describe English-language, Ireland and UK-centric answering.
Presence. The share of completed prompt runs in which the brand was mentioned, with an own brand evaluated only against its own prompt set. Position is the ordinal rank of the brand within the answer. Sentiment is model-extracted on a 0 to 100 scale. Figures were produced by read-only queries against the production and staging databases on the snapshot date of 22 August 2026.
Industry grouping. Industry is a free-text field enriched by a language model at onboarding, not a controlled taxonomy. We grouped the raw values lightly and published only industries carrying 100 or more own-brand evaluations.
Entity resolution. Brand mentions were originally detected by a word-boundary string matcher, which counted any answer containing the brand's name as a mention whether or not the answer was about that company. On 18 August 2026 we replaced it with a three-state verdict that can return uncertainty rather than forcing a match, and which is deliberately more willing to abstain than to claim. On the subset checked under the new method, 40 of 97 verdicts came back as a different entity. Each of the three corrections we have made to this layer has lowered the scores we show customers.
Replication. Staging is a separate environment with an overlapping but distinct brand set and independent runs. It is used strictly as a replication check on the shape of findings, never pooled with production numbers.