Back

Business & Company News 29/09/2026 06:16

REPORT REVEALS THE NEW COST OF AI: WHEN BUSINESSES ARE MISUNDERSTOOD OR MISSING FROM THE ANSWER

LEOPRD

Key Facts:

Nearly half (44%) of over 420,000 real-user prompts analysed involved AI supporting information, research, advice or judgement, with Australians more likely than UK or US users to ask AI to validate decisions or plan next steps — and AI validation usage rose 46% between late 2025 and mid-2026.
LEOPRD's analysis found that monitored brands were absent from 36% of AI responses to unbranded category questions, with commercial modelling associating this exclusion with approximately A$2.9 million in potential exposure for every A$100 million of Australian online retail spending - equating to around A$2.4 billion annually at market scale.
Being present in an AI answer did not guarantee being recommended; only 29% of responses made the monitored brand the primary recommendation, whilst 35% mentioned it but ultimately selected a competitor.
Researchers warn of cognitive surrender, with experimental studies finding participants followed AI advice 80% of the time even when it was incorrect, raising concerns about the authority AI can carry when answering consequential questions without sufficient scrutiny.
LEOPRD advises businesses to treat AI representation as a commercial outcome - connecting AI recommendation performance to qualified demand, returns, complaints, and churn - rather than viewing AI visibility solely as a marketing metric.

FROM SECOND OPINIONS TO SHOPPING DECISIONS: REPORT REVEALS THE NEW COST OF AI WHEN BUSINESSES ARE MISUNDERSTOOD OR MISSING FROM THE ANSWER

Australians are increasingly turning to AI for a second opinion as new research warns of ‘cognitive surrender’ and models A$2.9M in potential exclusion exposure for every A$100M of online retail spending

SYDNEY, 29 September 2026: Australian businesses face a new commercial risk as consumers increasingly use AI to help decide what to buy, who to trust and what to do next: the cost of AI making the wrong call - or leaving the business out of the answer altogether.

New research from AI visibility and reputation consultancy LEOPRD finds nearly half (44%) of 420,903 real-user prompts analysed from Prompt Cowboy involved AI supporting information, research, advice or judgement, while 23% explicitly asked AI to compare, evaluate, validate or recommend a decision.

16% of Australian prompts asked AI to check or confirm a decision, compared with 13% in the UK and 11% in the US. Australians were also the most likely to ask AI to help plan their next step (14%). AI validation rose 46% between late 2025 and mid-2026.

At the same time, separate analysis by LEOPRD found that when consumers asked AI unbranded questions about a category, the average monitored brand was absent from 36% of answers. LEOPRD's commercial modelling associates that exclusion with approximately A$2.9 million in potential exposure for every A$100 million of Australian online retail spending. Depending on assumptions around AI-shopping adoption and recommendation follow-through, the sensitivity range is A$2.2m-A$8.0m per A$100m. These are modelled exposure estimates, not measured revenue losses.

The findings form part of LEOPRD's new Reputation to Revenue 2026 report, which examines what happens as AI moves from information retrieval into research, judgement, recommendation and action.

To date, much of the debate around inaccurate AI answers has focused on hallucinations and misinformation. LEOPRD argues the business problem is becoming broader as AI moves closer to consequential decisions. When AI helps someone choose the wrong product or service, the cost doesn't necessarily stop with the customer.

Harding said: “When AI answers a question about a brand and gets it wrong, someone pays. The customer can waste money and time. The business can wear the return, complaint or support call and when a physical product is bought, shipped and returned unnecessarily, there is a cost to the planet too with unnecessary packaging, transport and waste.”

The problem isn't only AI getting facts wrong. Sometimes the user hasn't provided enough information, or AI simply doesn't have enough credible evidence about the business to make the right judgement and yet it can still give a confident answer.”

The report recommends businesses connect AI recommendation performance with outcomes including qualified demand, returns, complaints, service contacts and churn, rather than treating AI visibility as a standalone marketing metric.

FROM COGNITIVE OFFLOADING TO ‘COGNITIVE SURRENDER’

The growing commercial significance of AI recommendations comes as researchers examine how readily people defer to AI advice. Separate experimental research cited in the LEOPRD report found participants followed AI advice 80% of the time even when that advice was incorrect, compared with 93% when it was correct. The study involved experimental reasoning tasks and should not be interpreted as evidence that 80% of consumers follow incorrect shopping recommendations.

Professor Ben Newell, Director of the Institute for Climate Risk & Response and Professor of Behavioural Science at UNSW Sydney, said:

“People are asking AI to help frame what they need and then judge their options. The risk is that a confident answer can carry a veneer of authority, encouraging people to accept it without questioning the assumptions or evidence behind it. That goes beyond cognitive offloading. It becomes cognitive surrender.”

Newell has also warned that AI models can carry a “veneer of authority”, while often being designed to provide helpful, agreeable answers rather than challenge the assumptions behind a user's question.

Harding said the significance for businesses is not that AI has suddenly taken over consumer decision-making, but that it can influence which options people consider before they reach a brand's website, store or sales team.

“AI doesn't have to make the final decision to influence it. If it tells someone which three companies are worth considering, validates the option they're already leaning towards or gives them a reason to rule another one out, it has already influenced the commercial outcome.”

THE COST OF NOT MAKING THE ANSWER

LEOPRD separately analysed a sample of 31,200 AI responses from a broader dataset of more than 214,000, covering 27 companies across seven categories and eight AI platforms.

When consumers asked unbranded category questions, monitored brands appeared in an average 65% of responses and were absent from 36%.

Being known by AI did not mean being chosen. Across the study, 29% of answers made the monitored brand the primary recommendation, while 35% mentioned it but ultimately selected something else.

Harding said: “For years, businesses have worried about where they rank in Google but now there isn't necessarily a page of ten blue links where being fourth still gives you a chance. AI can synthesise the market, make comparisons and present a much smaller set of options directly in the answer.

If your business isn't in that answer, you may be removed from consideration before the customer even discovered you were an option.”

LEOPRD modelled the potential scale of that exclusion using Australian online retail spending, the proportion of shoppers using AI while shopping, the proportion reporting buying at least one AI-recommended product and the 36% exclusion rate observed in its monitored-brand study.

Its base case associates AI exclusion with $2.9m of potential exposure for every A$100m of Australian online retail spending. At market scale, the model equates to approximately $2.4 billion in annual Australian online retail spending potentially exposed to AI exclusion.

The figure does not represent revenue proven to have been lost because of AI. It estimates spending associated with the point at which AI use, recommendation follow-through and brand exclusion intersect.

AI IS MOVING CLOSER TO THE TRANSACTION

LEOPRD's prompt analysis found requests for AI to help with an action involving a vendor, service, or commercial process doubled between March and August 2026, from about six to 12 per 1,000 prompts. While still an emerging behaviour, these requests included preparing outreach, creating vendor requirements, structuring interactions with providers and taking steps within a purchasing workflow.

Harding said: “We're moving from ‘tell me about my options’, to ‘which option should I choose?’, to ‘is this choice right?’ and increasingly ‘now help me do something about it’.

As the distance between an AI answer and a commercial action gets shorter, businesses have less room to correct a poor recommendation after the fact. The question for leadership teams isn't simply ‘does ChatGPT know who we are?’ It's whether AI has enough accurate, credible evidence to understand when the business is the right answer - and what happens commercially when it doesn't.”

AI SEES THE BUSINESS, NOT THE ORG CHART

The report argues that responsibility for AI reputation cannot sit solely with marketing, communications or SEO because the evidence AI encounters is generated across the organisation.

Product capabilities and limitations, customer complaints and service resolution, regulation and safety, pricing and product information, external media coverage, reviews and leadership commentary can all contribute to the public evidence available for AI to interpret.

Harding said: “A marketing team can optimise content, but it can't optimise away a product problem, repeated customer complaints or an absence of independent evidence.

Businesses need to start treating what AI says about them as a commercial outcome: understand where they're being excluded, why another option is being chosen, fix the underlying evidence and then connect changes back to demand, returns, complaints, service contacts and churn.”

The full LEOPRD Reputation to Revenue 2026 report is available at report.leoprd.io.

FAST FACTS

44% of 420,903 user prompts analysed involved people using AI to find information, research options, seek advice or evaluate a choice.
25.3% of Australian money and finance prompts involved validation or active evaluation.
91% of Australian shopping prompts involved explicit decision support; 76% sought a recommendation.
Experimental research cited in the report found participants followed AI advice 80% of the time when it was incorrect and 93% when correct; these were experimental reasoning tasks, not shopping decisions.
The average monitored brand was absent from 36% of unbranded AI answers in LEOPRD's study.
Only 29% of answers made the monitored brand the primary recommendation.
LEOPRD's model associates exclusion with A$2.9m of potential exposure per A$100m of Australian online retail spending, with a sensitivity range of A$2.2m - A$8.0m. These are exposure estimates, not measured losses.
At Australian market scale, the base model equates to approximately A$2.4bn of annual online retail spending potentially exposed to AI exclusion.
Businesses are advised to connect AI recommendation performance to qualified demand, returns, complaints, service contacts and churn.

PROMPT COWBOY DEMO VIDEO
IMAGES

ENDS

For interviews and media enquiries contact:

Celia Harding | [email protected] | +61 428 229 455

Sally Douglas | [email protected] | +61 431 004 586

About LEOPRD

LEOPRD is an AI visibility and reputation advisory that helps brands understand and influence how they show up in AI to build reputation, drive recommendations and unlock commercial growth. LEOPRD combines expertise across PR and communications, search, digital and measurement to understand how AI platforms such as ChatGPT, Gemini, Copilot and Google AI Mode find, interpret and recommend brands.

Its work spans AI visibility and reputation audits, strategy, measurement and implementation, identifying where brands are missing, mentioned or recommended, the sources and signals shaping those outcomes, and what organisations can do to improve them.

Founded in 2025, LEOPRD works with brands across Australia, the UK and Europe. leoprd.io

About Prompt Cowboy

Prompt Cowboy helps people get better results from AI by turning rough instructions into clear, high-quality prompts for tools like ChatGPT and Claude. More than two million people use Prompt Cowboy, from professionals and business owners to everyday AI users. The idea is simple: you shouldn’t need to be a prompt engineer to communicate effectively with AI. Prompt Cowboy helps bridge the gap between what people have in their heads and what they actually tell AI, making powerful AI tools easier for anyone to use.

About the research

LEOPRD’s Reputation to Revenue Report 2026 combines consumer behaviour data, AI platform analysis and qualitative interviews.

LEOPRD analysed a random, anonymised sample of 420,903 prompts from Prompt Cowboy’s dataset of 10.4 million prompts, collected between November 2025 and August 2026. The sample covered Australia, the UK and the US and classified each prompt across 13 dimensions, including its purpose, sector, decision stage and the role AI was being asked to perform. No full prompt text or personally identifiable information was included.

LEOPRD separately analysed a sample of 31,200 AI responses from a broader monitoring programme of more than 214,000 responses. This research covered 27 brands, seven categories, eight AI platforms and two markets, Australia and Great Britain. It measured brand presence, consideration and primary preference, as well as the narratives, risks and sources associated with each brand.

The two datasets are separate and do not represent the same users or interactions. Qualitative interviews with communications, marketing and behavioural-science leaders provided further context for the findings.

Frequently asked questions

How are people using AI to make decisions?

People are using AI to gather information, conduct research, seek advice, compare options and test decisions. LEOPRD’s analysis of 420,903 anonymised and de-identified Prompt Cowboy prompts found that 44% asked AI to support some part of the user’s thinking, research or judgement. AI was used most heavily before a purchase. Pre-purchase research, comparison and validation accounted for 27% of prompts, compared with 1.3% for post-purchase support. This means people used AI around 20 times more often before deciding what to buy than to resolve a problem afterwards.

Can people trust the recommendations they receive from AI?

AI can provide useful advice, but its answers may be shaped by assumptions and evidence that are not immediately visible to the user. A short prompt may require the system to infer what the person means, which criteria matter, and how to assess competing options. LEOPRD’s research did not measure whether every answer was correct. However, a separate Wharton study involving 1,372 people found that participants followed AI advice 93% of the time when it was correct and 80% when it was wrong. Users should therefore check the sources, evidence and assumptions behind consequential AI recommendations.

How can businesses improve how AI represents and recommends their brand?

Businesses should begin by identifying the questions customers are likely to ask AI before making a decision. They can then test how different platforms answer those questions and whether the brand is accurately understood, considered and recommended. Important information should be clear, current and consistent across the company’s own channels. Claims should also be supported by credible independent sources, including reputable media, industry publications, reviews and expert commentary. Testing should be repeated over time because a single AI answer may not reflect the pattern customers encounter across platforms.

About us: