How Perplexity, ChatGPT And Gemini Pick Their Sources

Two implications follow regardless of which system you are studying. Being findable by the underlying search step is necessary, and being worth quoting once fetched is what decides whether you are used. Almost everything actionable sits in those two requirements.

Nineteen properties can show a real trend and cannot support a confident statement about the market. When that number is repeated without its sample size, as it usually is, it stops being evidence and becomes a slogan.

The Signals That Mean Nothing A rising composite visibility score with no methodology attached. The vendor controls both the number and the prompt set behind it, and it can improve without anything changing.

What Brands Usually Get Wrong in Response The instinctive response is to publish more brand content, which addresses none of the above. The second instinct is to try to displace the review site, which is not achievable and would not help if it were.

You will find your own category's pattern, which frequently contradicts the general one. Some industries are dominated by a single trade directory. Others are dominated by one forum. That specific finding is worth more than any general description of how these systems behave.

Search traffic includes everybody at every stage, including a large volume of people gathering background information with no intention of buying anything. Assistant referrals skip most of that, because the informational portion was satisfied before the click.

Before leaving, make sure you take the prompt set, the baseline archive and everything published. If those were not yours under the contract, that is a lesson for the next agreement rather than something to negotiate at the exit.

The Referral Growth Figure Is Weaker A widely shared statistic reporting several hundred percent growth in assistant referrals is worth handling more carefully still. Traced back, it rests on a sample of nineteen analytics properties.

If your category still gets meaningful traffic from those, a proposal scoped only to assistants will leave that work undone. Conversely, if somebody proposes an answer engine optimization programme and delivers only snippet optimisation, they are working on the older half of the definition.

What Ranking Does and Does Not Buy You Ranking still helps, because the retrieval step usually starts with a search. But it buys far less than people assume. Ahrefs examined 15,000 long-tail prompts across four assistants in July 2025 and found roughly 80 percent of cited pages did not rank for the original query at all, with about 12 percent in the top ten.

Gemini and Google Surfaces Closest to conventional search infrastructure, which has a practical consequence: work that improves your standing in Google search tends to carry over here more than it does elsewhere.

In practice it is used to mean roughly the same thing as generative engine optimization, occasionally with a stronger emphasis on training data and brand presence in the underlying corpus rather than on live retrieval.

Keep a dated note of what you observed each quarter, including behaviour that later turned out to be temporary. The value is not in the individual observations, most of which expire, but in noticing how fast they expire. A team that has watched three of its confident conclusions become wrong within a year develops the right amount of scepticism about the fourth.

The practical response to that uncertainty is to work on the things that are robust to it. Accessible pages, coherent identity, quotable writing and honest third party coverage have helped under every configuration observed so far, and they are the parts you would want anyway. homepage

Corroboration Beats Assertion The single clearest pattern in observed behaviour is that independent agreement outweighs self description. A claim made only on your own site is treated as a claim. The same claim appearing on a review platform, in a trade publication and in a forum thread is treated as a fact about the world.

Run a commercial prompt in almost any category and look at what gets cited. Review platforms, roundups and comparison sites appear first and most often, and the brands being discussed appear well down the list if at all.

How to Test Rather Than Trust Everything above is a starting hypothesis. Run twenty prompts in your own category across all three, from signed out sessions, recording the mode and the date, and count the cited domains for each.

The defensible version states the mechanism, cites the available evidence with its sample sizes, presents your own segmented data however thin, and is explicit that most of the channel's value is not measurable through referrals at all.

What the Evidence Actually Is The figure quoted most often comes from Opollo, which reported assistant referred traffic converting at 14.2 percent against 2.8 percent from conventional search. The sample was 312 business to business brands, attributed through UTM parameters, covering the third quarter of 2024 through the first quarter of 2025.