Model visibility in ChatGPT shifts immediate by immediate. Two associated questions in the identical purchaser’s analysis can floor utterly totally different manufacturers. Observe just one, and also you’d assume you are profitable. Look throughout the subject, and the image adjustments.
The query in AI search is now not “do I rank for this key phrase or immediate?” It’s “when the AI solutions questions throughout the subjects key to my enterprise, does my model maintain displaying up?”
To learn how manufacturers win and lose visibility throughout subjects, we partnered with Kevin Indig to check 1,094 topic areas in ChatGPT.
Here is what we discovered.
Key takeaways
- Successful one immediate in ChatGPT just isn’t the identical as proudly owning a subject. Actual matter possession requires a model to seem throughout at the least 4 of 5 associated prompts, with a 5-percentage-point lead over the runner-up.
- Most subjects in ChatGPT are nonetheless contestable. Solely 15.2% of the 1,094 classes we analyzed had a transparent proprietor. 85% stay open to opponents that may construct constant visibility.
- Broad search engine optimization indicators don’t constantly predict matter possession. Area-level metrics like Authority Rating and natural visitors correlate with possession solely about half the time. The aggressive image solely turns into clear on the matter degree.
- As soon as a model owns a subject in ChatGPT, it tends to carry on to it. However solely with a large sufficient margin. Homeowners with a 5-percentage-point lead retained first place in 90% of month-over-month comparisons. Slim leads change palms incessantly.
Methodology
We partnered with Kevin Indig and Development Memo to research information from the Semrush AI Visibility Toolkit.
The dataset covers 1,094 US classes, tracked month-to-month in ChatGPT from January via June 2026.
Every class is a topical cluster of 5 consultant prompts overlaying the questions a purchaser would possibly ask in that topic space (definition, comparability, options, use case, shopping for query).
Consider a class as a subject like “payroll software program,” “small enterprise banking,” or “greatest pillows for neck ache,” and the 5 prompts as the usual questions somebody would ask about every.
Scope of the info:
- Over 220,000 domains
- Over 50,000 manufacturers
- Over 600,000 citations
- Over 220,000 URLs
For every immediate, we tracked which manufacturers ChatGPT talked about within the reply and which sources it cited as hyperlinks.
Definitions
- Class proprietor: the model with the very best share of mentions, named in at the least 4 of 5 prompts, with at the least a five-percentage-point lead over the runner-up.
- Rising chief: the most-mentioned model seems in at the least three prompts however doesn’t meet the proprietor threshold.
- Unsettled class: no model seems in at the least three prompts.
1. Most classes are nonetheless up for grabs in ChatGPT”
Solely 15% of the AI search classes we analyzed have a transparent model winner. Within the different 85%, no single model reveals up constantly throughout the subject’s associated prompts. A model would possibly seem in a single query and be lacking from the following.
Right here’s the complete breakdown of the 1,094 classes we analyzed in ChatGPT:
- 15.2% have a transparent proprietor (one model seems in at the least 4 of 5 prompts with a 5-point lead)
- 31.2% have an rising chief (one model seems in at the least 3 prompts, however with out the required lead)
- 53.7% are unsettled (no model seems in even 3 of 5 prompts)
Here is the counterintuitive half: You’d count on the most important subjects, those folks ask ChatGPT about most frequently, to be dominated by established manufacturers like they’re on Google.
Greater search quantity often means stronger competitors. In ChatGPT, it is the other
We break up the 1,094 classes into two teams by AI search demand: the highest half (the subjects folks ask about most) and the underside half (those they ask about much less). Solely 11ю3% of the top-half subjects had a transparent model proprietor. Within the backside half, 19% did.
And that high half accounts for 98% of the AI search quantity in our pattern, so nearly each query we tracked in ChatGPT lands in a subject with out a clear winner.
The takeaway: profitable one immediate would not make you the model ChatGPT associates with a subject. Exhibiting up constantly throughout all of the associated questions does. Within the majority of subjects we analyzed, no model managed to do this but, which is the chance.
2. search engine optimization metrics do not at all times predict who wins a subject in ChatGPT
If a model has sturdy search engine optimization fundamentals (excessive natural visitors, excessive branded search quantity, a strong Authority Rating), you’d count on it to win in AI search too.
We examined this by evaluating the subject proprietor in opposition to the runner-up on three domain-level metrics: branded search quantity, natural visitors, and Authority Rating.
The outcomes solely line up with possession about half the time. Solely branded search quantity was a statistically vital differentiator, and even then, the sting was modest.
- Homeowners had larger branded search quantity in 55.7% of pairs
- Homeowners had larger natural visitors in 48.4% of pairs
- Homeowners had a better Authority Rating in 52.5% of pairs
In almost half the subjects, the runner-up had at the least one among these benefits and nonetheless misplaced. No matter decides who owns a subject in AI search is not captured by broad search engine optimization power alone.
“Deal with that as a speculation, not a discovering. What the info really reveals is: conventional search engine optimization metrics aren’t sufficient to clarify who owns a subject. Whereas they play their function, there’s extra to it.”
Robust search engine optimization helps you present up, however as soon as a number of manufacturers are within the operating, deep protection of the precise matter is what turns you from runner-up to proprietor. In case you’re sturdy in Google however dropping matter authority in ChatGPT, the repair is not extra homepage backlinks. It is constructing a deeper library of content material and getting related mentions on that particular topic on high of the search engine optimization basis.
3. When you personal a subject in ChatGPT, you have a tendency to carry it. However provided that your lead is large.
When a model has a snug lead in a subject in ChatGPT, they nearly at all times maintain it. When the lead is slim, they usually lose it.
Clear class house owners we analyzed stayed on high in 90.4% of month-over-month comparisons. Rising leaders and unsettled classes modified palms way more usually, with the highest model switching in 1,950 of 5,470 comparisons.
- Subjects the place the chief flipped: median lead of 1.3 proportion factors
- Subjects the place the chief held: median lead of two.9 proportion factors
Management modified palms often. Nevertheless, these shifts have been concentrated in classes the place the preliminary lead was small.
Right here’s an instance the place a transparent chief was changed by the earlier month’s runner-up:
Why the switches occur is not one thing this information can reply. They could mirror shifts in ChatGPT’s sources, adjustments in the way it generates solutions, or actual shifts in a model’s underlying relevance. What the info reveals is a less complicated sample — margin issues greater than route.
The takeaway: development would not equal security. In case you lead a subject with solely a slim margin, maintain pushing content material and protection. A competitor rising sooner can nonetheless overtake you. In case you’re a challenger in a subject the place the present chief has a slim lead, that is the second to behave. As soon as they consolidate a much bigger margin, the subject will get a lot tougher to maneuver.
Find out how to begin profitable on the matter degree
The manufacturers profitable in AI search function on the class degree. They decide a targeted set of subjects and construct the protection, mentions, and third-party authority indicators ChatGPT depends on inside every one.
This takes time to construct, nevertheless it compounds. When you grow to be the model ChatGPT associates with a class, there’s a great likelihood you’ll be able to maintain it.
Particular person prompts are nonetheless value watching as a sign. However your topical authority determines whether or not ChatGPT will floor you when a buyer asks one thing new in that class.
Here is the best way to get began:
- Determine the topical areas that matter for your online business. Begin from what your prospects care about and the place you compete. The classes of questions your patrons ask after they’re evaluating one thing like what you promote are the subjects value constructing round.
- Analyze your present topical authority. Semrush’s AI Visibility Toolkit already helps you to see which subjects you are influencing in AI search and the place you sit inside every one. Begin there to get an sincere baseline of the place your authority is actual, the place it is skinny, and the place opponents are forward.
- Focus your efforts. Making an attempt to construct authority throughout each matter without delay dilutes all the pieces. Choose two or three related classes the place you have already got some sign, present content material, and an actual product angle. That is your first guess.
- Develop your topical authority. Topical presence comes from three issues working collectively: broad content material protection throughout the questions inside a class, related mentions within the sources ChatGPT attracts from (Reddit, evaluate websites, trade publications, professional commentary), and robust entity indicators (structured information, constant model naming, third-party citations that tie your model to the subject).
