How topical authority spreads (and where it doesn’t) in ChatGPT [Study]

In our earlier examine, we discovered that AI visibility is a topic-level sport. Successful a single immediate doesn’t robotically make you the model ChatGPT connects with a class. 

This examine asks the following query: when you’ve constructed that sort of authority in a single class, does it carry over into associated ones?

In case your model turns into a standard reply in “small enterprise banking,” does that make it extra more likely to present up in “enterprise bank cards” or “small enterprise loans”?

The brief reply: it is determined by whether or not you’re measuring citations or model mentions. Within the manufacturers we analyzed, citations tended to unfold extra readily throughout classes than model mentions did. And the model’s experience and depth in a class formed how far its authority appeared to journey.

Key takeaways

• Among the many manufacturers we analyzed, a website may very well be cited in nearly any class if its content material met the sort of supply standards ChatGPT depends on (relevance, authority, freshness). Getting talked about by identify in ChatGPT tended to occur largely in classes near a model’s core experience.
• For model mentions particularly, depth in a class tended to matter greater than breadth. Manufacturers that constantly appeared throughout a class’s prompts in ChatGPT had been those related to profitable growth into new classes. Displaying up as soon as in 20 classes didn’t seem to construct a better brand-mention share.
• Business formed the sample. Within the manufacturers we analyzed, finance and actual property noticed broader protection related to extra citations. Authorized and healthcare set a better bar: even full immediate protection and repeat citations didn’t reliably translate into brand-mention share.
• Topical authority tended to provide manufacturers a stronger place to begin when increasing into adjoining classes, not a free cross into unrelated ones. Growth labored greatest when the following class was genuinely shut to 1 the model already answered properly. 

Methodology

We partnered with Kevin Indig and Growth Memo once more to investigate knowledge from the Semrush AI Visibility Toolkit.

The dataset covers 1,094 US classes tracked month-to-month in ChatGPT from January via June 2026, with 5 immediate variants per class.

It contains 283,215 domain-category quotation observations and 76,493 brand-mention observations throughout January via Might, plus 45,578 category-expansion appearances by 1,458 mapped model entities.

Definitions

Two alerts are tracked individually all through the examine:

• Citations are the sources ChatGPT hyperlinks to as proof in its solutions
• Model point outs are the model names ChatGPT contains immediately within the response textual content (which can be impartial, constructive, or unfavourable)

For the relatedness evaluation, a model’s “core experience” is outlined as any class the place the model had already appeared in no less than three of 5 prompts earlier than the target-month look.

“Class closeness” is calculated utilizing semantic similarity between the goal class’s prompts and the model’s prior professional classes.

1. The bar for being talked about is increased than the bar for being cited

When a model in our examine appeared in a topical class in ChatGPT, whether or not it was simply cited or additionally talked about by identify depended closely on how associated the class was to its core experience.

In classes near the model’s core experience, manufacturers had been cited as a supply in 74% of appearances, talked about in 44%, and each cited and talked about in 34%. In classes distant from the model’s core experience, these shares dropped to 50% cited, 25% talked about and 9% each cited and talked about.

Being cited as a supply held up moderately properly throughout the gap (50% distant vs. 74% shut). However being named alongside a quotation was strongly tied to relatedness — cited-and-named appearances had been almost 4x increased in shut classes than in distant ones.

The sample was even sharper in the event you seemed solely at citations. Of the citations that occurred within the least associated classes, simply 18% additionally included a named model. In probably the most related, 46% did. 

Associated classes had been greater than twice as more likely to flip a quotation right into a named-brand point out on this pattern.

The takeaway: match your technique to your objective. Publishing broadly can construct quotation presence, and that seems to work even in classes far out of your core. However if you would like ChatGPT to call your model when a buyer asks a query, focusing your effort on classes near what you’re already recognized for tends to repay extra.

2. Depth issues greater than breadth in ChatGPT for model recognition

In our examine, a model might present up as soon as in 20 classes and personal none of them. Depth in a single class was related to a better brand-mention share. Breadth throughout many classes, with out depth, was not.

Quotation share, nonetheless, responded to protection both means. A model that confirmed up in even one in all a class’s 5 prompts tended to achieve some quotation share. A model that confirmed up in all 5 gained extra. Within the pattern, there was no apparent spread-thin penalty for citations.

Model mentions behaved in another way. A model that appeared in solely one in all a class’s 5 prompts was related to a drop in point out share. That penalty didn’t disappear till the model reached no less than three of 5 prompts in a class. Solely after that did breadth cease hurting model mentions.

“Being cited in lots of classes doesn’t present a spread-thin impact on this pattern. However model recognition works in another way. The classes the place a model constantly earns named mentions are those from which it could possibly credibly increase.”

Kevin Indig, Founding father of Progress Memo

The takeaway: equally, prioritize depth in a couple of classes over presence in lots of. The info doesn’t assist the concept that a web site can serve limitless classes, nevertheless it additionally doesn’t assist a common cap. What it does assist is an easy rule: earn repeat presence in a class earlier than treating it as a launching pad for the following one.

3. Business shapes how far authority spreads in ChatGPT

In response to our knowledge, how protection breadth impacts a model’s presence in ChatGPT isn’t the identical throughout industries.

In finance and actual property, wider protection tended to carry extra quotation leverage. A model that confirmed up in all 5 prompts in a class was related to stronger quotation share than one which confirmed up in just one.

That is to not say any finance or actual property model can dominate each adjoining class. It simply implies that in these industries, spreading protection throughout many classes did not seem to harm a model’s quotation share.

Authorized and healthcare set a better bar. Their quotation beneficial properties from broader protection had been weaker than in finance or actual property. And for model mentions, protection didn’t translate: even at full five-prompt protection, point out share nonetheless trended unfavourable in these fields. Being extensively cited didn’t translate into being named.

The best way to plan topical class growth in AI search

To increase efficiently in AI search, do not chase each content material matter. Begin with the areas the place you are already acknowledged, then increase outward in small, deliberate steps, specializing in what strikes the needle to your model.

Right here’s learn how to plan an growth transfer:

1. Map the place you have already got authority. Listing the classes the place your model constantly earns each citations and named mentions throughout a lot of the associated class prompts. These are your anchor classes, not each class the place you’ve appeared as soon as.

Begin by checking the matters you already seem for in Semrush’s Visibility Overview tool:

2. Choose goal classes which are genuinely near your core. On this examine, closeness was measured by semantic similarity between the anchor and the goal class. In follow, that always means in search of classes the place the client, downside, product, and shopping for determination look comparable and are immediately related to your small business.

Open the Subject Opportunities tab to see what it’s best to deal with subsequent:

3. Give attention to what earns model mentions. To earn mentions in a brand new class, deal with clear model positioning throughout your content material, third-party protection from credible sources (Reddit, assessment websites, business publications, professional commentary), and proof of experience (creator credentials, major analysis, actual product particulars). These weren’t examined immediately on this examine, however they’re frequent alerts that are likely to impression AI visibility.