Brands are failing to become AI’s first choice in seven out of ten unbranded answers, despite often being known by the model, according to new research from AI visibility and reputation advisory LEOPRD.
Analysis of a sample of 31,200 AI responses, drawn from more than 214,000 monitored-brand responses, found brands were missing from 36 per cent of unbranded answers, mentioned but not preferred in 35 per cent, and most wanted, or AI’s primary recommendation, in just 29 per cent.
Separate analysis of 420,903 anonymised and de-identified prompts in the study found people used AI around 20 times more often for research, comparison and validation before a purchase than for post-purchase support, 27 per cent of prompts versus one per cent.
Additional analysis of the Australian data found 91 per cent of shopping prompts analysed involved explicit decision support, including comparison, evaluation, validation or recommendation-seeking, while 76 per cent specifically sought a recommendation.
That puts AI firmly inside the customer journey while options and shortlists are still being formed, making the difference between being mentioned and being recommended increasingly commercially significant.
High visibility scores also hid significant recommendation gaps. Among the fintech brands tested, monitored brands appeared in 92 per cent of relevant answers but were first choice in only 32 per cent of answers in which they appeared. Among the travel and transport brands tested, monitored brands appeared in 90 per cent, but were first choice when present just 28 per cent of the time.
“Being found isn’t the same as being selected. If ChatGPT mentions you alongside four competitors and then tells the customer somebody else is the better choice, you’ve achieved visibility, but you’re unlikely to be chosen,” said Celia Harding, founder of LEOPRD.
Both analyses form part of research conducted for LEOPRD’s new Reputation to Revenue 2026 report, conducted in partnership with Prompt Cowboy. Prompt Cowboy is a free platform that turns a lazy prompt into a clearer and more detailed brief for AI tools such as ChatGPT and Claude. LEOPRD analysed anonymised, de-identified Prompt Cowboy data to understand how people are using AI across Australia, the UK and US.
AI visibility facts: 36 per cent mentioned the monitored brand was absent from more than one in three unbranded AI answers; 35 per cent said the brand appeared, but AI endorsed someone else; the brand was AI’s primary recommendation in fewer than one in three answers; 91 per cent of Australian shopping prompts analysed involved explicit decision support, with 76 per cent seeking a recommendation; AI was used 20 times more often before a purchase than after: 27 per cent of prompts, versus 1 per cent for post-purchase support.
Across Australia and Great Britain, AI drew on broadly similar mixes of source categories, while the local publishers and review systems differed; the overall recommendation challenge did not.
The cost of being absent
When consumers asked AI an unbranded category question, the average monitored brand was absent from 36 per cent of answers. LEOPRD’s commercial modelling associates that exclusion with approximately $2.9m in potential exposure for every $100m of Australian online retail spending. The figure represents modelled commercial exposure, not measured revenue loss.
“If you’re not in the answer, you don’t get the chance to be chosen. As AI moves further into research, comparison and purchase decisions, being absent increasingly means being absent from consideration,” said Harding.
Being left out of an AI answer is one commercial risk. Being included in a poor answer creates another. When AI is part of a purchase or service decision, incomplete or unreliable information can lead customers towards the wrong product, service or next step. The cost can then move downstream to the business through returns, complaints, support calls and churn.
The report points to customer service as one example. Gartner research cited by LEOPRD found only 14 per cent of customer-service issues were fully resolved through self-service, while the median cost of an assisted contact was USD $13.50 compared with USD $1.84 for self-service.
In retail, the same principle applies when poor decision support contributes to a wrong-fit purchase: the customer loses time, the business absorbs the cost of returns and handling, and physical returns create additional transport, packaging, processing and waste.
Answers where the monitored brand was included but not chosen averaged 9.2 citations, compared with 8.4 when it was AI’s primary recommendation. The difference was not the amount of evidence, but what that evidence said about the brand and whether it gave AI a clear reason to recommend it. Eighty-six per cent of the evidence came from external channels.
Across the study, 86 per cent of citations came from sources outside brands’ own properties, including media coverage, comparison sites, reviews, communities, research and institutional sources.
The evidence AI relied on also differed by category. Forty-one per cent of cited sources in captured travel primary-recommendation answers were editorial, while B2B software recommendations leaned heavily on commercial and comparison sources. Health and care relied more heavily on owned information and research than many other categories.
Canva appeared in 86 per cent of relevant answers and was first choice in 99.7 per cent of answers in which it appeared. Octopus Energy appeared in 97 per cent and was first choice in 92.9 per cent when present. Yet the broad mix of commercial, editorial and review sources behind strong performers was strikingly similar to brands that appeared frequently but were rarely chosen. What differed was what those sources said about the brand and whether they gave AI a clear reason to recommend it.
Only 40 per cent of repeated-test groups returned the same recommendation outcome throughout, even when LEOPRD tested the same question, platform and market repeatedly.
Harding said: “One answer is an observation, your AI reputation is a pattern. It’s these patterns across prompts, platforms and time that should guide AI strategy.”
AI draws on evidence created across the business, from product information and customer experience to reviews, media coverage, pricing, leadership and reputation.
“This is not an SEO play, where you’re bidding for words. It’s pushing everyone back up to what we used to do with marketing: do we know who we are, do we understand our customer, and how do those two come together? This spans pretty much the whole length of marketing, from brand, to customer insights, to content. And then there’s PR and communications, which also needs to be part of that whole, rather than marketing sitting here and PR sitting over there, never speaking. The two are now so interconnected it almost becomes one,” said Jodie Sangster, co-founder of the Australian Centre for AI in Marketing (ACAM).
More findings from Reputation to Revenue 2026 will be presented at the HumAIn conference on 13-14 October.
