Guide · 3 min read
How AI assistants name brands in their answers
What the stored answers show about naming patterns across ChatGPT, Gemini, Perplexity and Claude, without guessing at their internal workings.
Observing outputs, not algorithms
AI assistants do not publish the rules that decide which brands appear in an answer. Anyone who claims to know those rules precisely is guessing. The Index takes a narrower and more verifiable approach: it asks the same questions every week and records what comes back.
From those stored answers it is possible to describe patterns. Some brands are named by every assistant in most answers. Others appear often in one assistant and rarely in another. Some are named mainly when a question mentions a particular need, such as a price range or a life stage. These are observations about outputs, and the guide keeps to them.
What a typical answer looks like
Most recommendation answers name between three and eight brands, often as a short list with a sentence on each. Some begin with a single suggestion and then add alternatives. Others list comparison sites or public bodies alongside companies, which is why category pages separate brands from comparison sites, publishers and other entities.
The Index records each named entity, the order in which it appeared and, where the assistant cited web sources, which domains were cited. It does not score tone. A brand mentioned as a cautionary example is still counted as named, which is one reason visibility should be read as presence rather than endorsement.
Answers with web search switched on often reflect what is currently written about a category online. That can make figures move after a widely covered product launch, a price change or a regulatory story, though the Index does not attribute any individual change to a specific cause.
Why the assistants disagree
The four measured assistants are built by different companies, use different models and retrieve different sources. It would be surprising if they named the same brands at the same rates, and the data confirms that they often do not.
Category pages show visibility for each assistant separately, and the consensus view lists brands where the recorded rates differ most. A wide spread is not an error. It tells you that a brand's visibility depends heavily on which assistant a person happens to use.
For a brand, that matters in practice. Strong overall visibility built on one assistant is more fragile than the same figure spread evenly. Equally, a brand that is weak overall but strong on one assistant may already be doing something that assistant's retrieval picks up.
Model changes
Assistants update their default models from time to time. When that happens, naming patterns can shift. The Index logs every default model change it detects and marks it on history charts so that a step change can be read in context.
A model change is not proof that the new model caused a movement. Other things change in the same weeks. The marker is there so a reader can see that the measurement conditions changed, and treat the week either side with appropriate care.
Using this as a brand
Start with your category page and look at the per-assistant columns rather than the headline figure alone. Note which assistants name you and which rarely do. Then look at the brands that are named more often by the assistants where you are weak, and at the sources those answers cite.
That comparison will not reveal a secret formula, because none is published. It does give you a concrete, repeatable picture of where you appear and where you do not, measured the same way every week, which is a sounder basis for decisions than anecdote or a single manual test.
Live evidence · In the week to 13 November 2026 (week 47, published 16 November)
Each assistant's most-named brand
For a sample of published categories, the brand each assistant named most often in the measured window.
| Category | ChatGPT | Gemini | Perplexity | Claude |
|---|---|---|---|---|
| Banking and current accounts | NatWest (38.5%) | NatWest (53.5%) | HSBC (51.5%) | NatWest (45.5%) |
| Investing and wealth | Interactive Investor (39.0%) | AJ Bell (64.5%) | Interactive Investor (37.0%) | AJ Bell (80.0%) |
| Mortgages and personal finance | Tesco Bank (22.0%) | Santander (30.5%) | HSBC (25.5%) | HSBC (25.0%) |
| Business finance and payments | Square (22.5%) | Xero (37.0%) | Square (21.5%) | Xero (28.5%) |
| Cars and motoring | Kia (37.5%) | Hyundai (40.5%) | Carwow (45.0%) | Hyundai (36.0%) |
| Fashion and clothing | Marks & Spencer (40.5%) | Marks & Spencer (37.0%) | Marks & Spencer (55.0%) | Marks & Spencer (40.0%) |
Full definitions and limitations are in the methodology.