New Captify research finds around 70% of AI responses centre on a single primary brand rather than a list of options. LLMs also describe brands more warmly than the wording in the prompt. Being the one brand the AI names is the new shelf position.
Search gave you a page of ten blue links to compete on. AI gives you one name. Being second is being nowhere.
New research from Captify lands the point that should worry every marketer relying on being one option among many. Around 70% of AI responses centre on a single primary brand rather than presenting a spread of alternatives. The AI does not hand the customer a shortlist. It hands them an answer.
The study, titled Prompt to Purchase, also found that LLMs tend to describe brands more warmly than the wording people use in their prompts. That optimism bias strengthens brand perception and nudges purchase intent before the person ever reaches the open web. Captify puts the tipping point at around five to six prompts, the window where a shopper moves from exploring to deciding.
This is a different game to search. Ranking fourth on Google still put you in front of people. Being the second brand an AI considers means you were not mentioned at all. The distribution is winner-take-most, and the winner is decided inside a model you do not control.
Why it matters
Australian businesses have spent two decades learning to win a page of results. That skill is depreciating. When the answer is a single recommendation, the job shifts from ranking to being the brand the model has the most, and the most positive, evidence about. That is built on reviews, third-party mentions, consistent entity data and being cited across the sources these models trust. The brands that own their category conversation offline and across the web are the ones the AI keeps naming.
Share of AI responses that recommend a single primary brand rather than a list of options. Source: Captify.
What to do about it
Test your own category. Ask ChatGPT, Gemini and the AI Overview what they recommend in your space. See if you are named.
Audit your citation footprint. Reviews, directories, comparison sites and press are the evidence the models read. Thin evidence means a thin chance of being the pick.
Fix your entity data. Consistent name, category and details across the web make you easier for a model to recommend with confidence.
Build genuine third-party proof. The brand with more credible mentions wins the single slot, not the one shouting loudest on its own site.
The shelf just shrank from ten spots to one. Work out today whether you are on it.