The comfortable assumption in most boardrooms is that AI answers will earn more trust as people come to rely on them more. Familiarity breeds confidence. The reality is running the other way. Consumer trust in AI search fell from 82% in 2025 to 54% in 2026, a 28-point drop in a single year, even as buyers fold these tools deeper into how they check out a company, according to research from Fractl and Search Engine Land (a publication owned by Semrush). That gap is the whole problem. An AI system now compresses your entire content footprint into a two-paragraph verdict before a buyer, a journalist or a board member reaches anything you control. This briefing sets out why brand reputation in AI answers has become a standing board-level risk, and what a comms leader can actually stand up in response.
The compression problem: your brand, summarised before the first touchpoint
Reputation used to build gradually. Someone read a piece of coverage, browsed your site, asked a colleague, then formed a view over several encounters, which gave you time to shape it. That process is being collapsed into a single answer. As Neil Patel's team argues, an AI-generated summary can now stand in for all of those touchpoints, and it does not privilege the content you own. It pulls from your site, press coverage, review platforms, forums and complaint boards, then weights them in ways that are not always intuitive.
Think of it like a credit score for your reputation. A prospect never sees the underlying file, only the compressed verdict, assembled from records you did not choose and cannot directly edit. The score arrives before the conversation does. The first impression of your brand is increasingly formed in a system you do not own, from sources you may never have audited.
Trust is falling while reliance keeps growing
This is what widens the risk rather than narrowing it. A channel people distrust but keep using is more dangerous than one they abandon, because buyers act on the verdict, then discount your correction. The Fractl and Search Engine Land data shows the skeptic camp grew sixfold in a year, and buyers now cross-check an average of 2.4 platforms before validating a purchase. Doubt has not stopped people using AI answers. It has just made them harder to reassure.
The Australian picture rhymes with the global one. A Q1 2026 survey of 1,200 Australians, reported by Business News Australia, found 66% do not trust the accuracy of AI-powered answers, and 81.5% want AI-generated images and video clearly labelled [AU]. On the supply side, disclosure is the exception. The Fractl research found that while 84% to 91% of consumers want AI labelling across formats, only about 20% of organisations always disclose their AI use and a third never do. Australia's own privacy regulator sees the same pattern in government: reviewing 23 entities, the OAIC found that only four disclosed their automated decision-making, and none had published guidelines [AU]. Trust is scarce, and the disclosure that might rebuild it is rare.
Partial truths and the most-repeated-wins dynamic
The failure mode boards picture is fabrication. The more common one is quieter and harder to fight. Neil Patel's analysis makes the point that the real risk is partial truths: accurate statements pulled out of context, outdated positions that were once correct, nuanced stances flattened into something you no longer hold. These are more corrosive than outright falsehoods because they are harder to dispute. And once an AI system assembles a narrative, it reinforces it every time someone asks a related question. The most repeated claim, not the most accurate one, is what rises to the top.
Regulators are starting to name this directly. The ACCC's December 2025 industry snapshot lists, among emerging risks, AI chatbots producing false but authoritative-sounding statements in response to user queries, and flags manipulative or false practices in digital markets as a 2026-27 enforcement priority [AU]. The advertising side is moving too: Australia's Ad Standards has flagged AI's potential to mislead as a growing area of public concern, and the industry body AANA is reviewing its Code of Ethics over whether existing misleading-and-deceptive rules stretch to generative AI [AU]. There is no blanket AI-disclosure rule in Australian advertising yet, but the direction of travel is clear enough to plan around.
Why the quick fixes fail, including LLMS.txt
The temptation is to reach for a technical patch. The most popular one has just been ruled out. Google has confirmed that LLMS.txt files neither help nor hurt search rankings and are not used by Search at all. A file you drop on your server does not govern what an AI says about you. Neither does a one-off audit, because the inputs keep changing as the web changes.
What replaces the patch is governance treated as infrastructure rather than administration: the standing discipline that keeps your brand signals clear enough to survive compression. This is not a marketing footnote. It is a governance question, and senior authorities are framing it that way. In Australia, the NSW Chief Justice has warned that directors cannot blindly adopt the recommendations of an AI system, and that its fluent, confident output can create a false confidence that erodes proper debate [AU]. In the UK, regulators including the ICO, CMA and FCA increasingly expect explainable logic behind automated decisions, and commentary on responsible AI puts the reputational cost above the legal one: "trust collapses fast" when a business cannot explain what its systems did [UK]. Worth noting that piece is from mid-2025, so read it as framing rather than a fresh development.
A caution on vocabulary. The label for optimising toward AI answers, generative engine optimisation or GEO, is contested. Google's own position is that good SEO is good GEO, so treat GEO as a useful lens, not a settled discipline with guaranteed levers.
What a comms leader should stand up
The instinct to run one audit and move on misreads the environment. Neil Patel's team recommends monitoring on a standing cadence, not quarterly, because a viral post, new coverage or a competitor's messaging shift can change what an AI says about you within weeks. So the first thing to stand up is a recurring monitoring rhythm: prompt the major tools with the questions a buyer, investor or journalist would ask, capture the outputs, and trace each narrative back to the sources feeding it.
The second is disclosure, framed as a brand-trust signal rather than a compliance checkbox. When most organisations either never disclose or treat labelling as legal cover, the ones that disclose plainly stand out on exactly the axis buyers say they care about.
The third is third-party authority, because AI systems weight it heavily. The Fractl research, drawing on Ahrefs analysis of roughly 75,000 brands, found branded web mentions and YouTube impressions correlated most strongly with AI visibility (around 0.50 to 0.74 on the Spearman scale), while backlink count and ad spend sat in the weakest tier, below 0.30. Earned media, analyst coverage and credible reviews are what the models trust. Underneath all three sits message consistency: Search Engine Land's analysis of retrieval versus citation argues that a consistent brand story across every channel is what lets machines understand who you serve.
None of this yet appears on a board agenda under the heading "AI-search reputation." That framing is our argument, not a line item directors have already adopted. But the adjacent ground is firmly on the agenda. In Australia, the AICD's 1H26 Director Sentiment Index, based on 828 directors surveyed by Roy Morgan, records concern about AI climbing, with over 80% expecting more AI implementation in the year ahead and potential AI misuse named among major risks [AU]. The AICD and UTS Human Technology Institute have gone further, with director guidance on AI governance that frames AI reputational damage as a board duty rather than a peripheral IT matter [AU]. The UK sits in the same place: a survey of more than 600 UK directors found half naming a lack of trust in AI outcomes as one of their biggest concerns, and the Institute of Directors' board-meeting guidance, drawing on roughly 700 directors and business leaders, explicitly ties AI governance to reputation [UK]. The question a comms leader can put in a board paper is simple: what will AI tell someone about us, and is it accurate?
Key takeaways
- Treat brand reputation in AI answers as a standing function, not a project. The inputs change continuously, so a one-off audit or an LLMS.txt file will not hold, as Google has confirmed the file has no ranking effect.
- Assume the compressed AI verdict reaches your buyer before your owned content does, and that consumer trust in that verdict is falling (82% to 54% in a year), which makes accuracy and disclosure more urgent, not less.
- Watch for partial truths and the most-repeated-wins dynamic rather than outright fabrication. Australian regulators, including the ACCC, are already naming authoritative-but-wrong AI output as a risk.
- Invest where the models place trust: earned media, analyst coverage and credible third-party mentions, backed by a consistent brand message across every channel.
- Give the board language it can use. AI risk and AI reputational harm are already on the AICD and IoD agendas, so connect your monitoring to a governance frame directors already recognise.
Frequently asked questions
What is brand reputation in AI answers?
It is the impression an AI system gives when someone asks about your company, assembled from your whole content footprint rather than only the content you own. Because AI compresses that footprint into a short verdict before any owned touchpoint, that summary increasingly shapes first impressions among buyers, journalists and even directors.
Is trust in AI search actually falling?
Yes, at least on the available data. Fractl and Search Engine Land research found consumer trust in AI search fell from 82% in 2025 to 54% in 2026, with the skeptic camp growing sixfold, and an Australian survey of 1,200 people found 66% do not trust the accuracy of AI answers. Reliance is still rising, which is what makes the falling trust a reputational risk rather than a reprieve.
Will an LLMS.txt file fix how AI describes my brand?
No. Google has confirmed that LLMS.txt files neither help nor hurt search rankings and are not used by Search. There is no single file or one-off audit that governs what AI says about you, because the sources feeding those answers change constantly.
Is this really a board-level issue or a marketing one?
It sits with both, and director bodies are already treating AI risk as a governance concern. The AICD's 1H26 Director Sentiment Index shows AI concern climbing among 828 directors surveyed, and the Institute of Directors' guidance in the UK links AI governance to reputation, though neither has yet named AI-search reputation as a specific agenda item.
What should a comms team do first?
Stand up a recurring monitoring cadence: prompt the major AI tools with the questions your buyers, investors and journalists ask, capture the outputs, and trace each narrative back to its sources. From there, prioritise disclosure as a trust signal and invest in the earned media and third-party authority that AI systems weight most heavily.
Where this leaves you
If AI reputation is now a live, unowned narrative shaped by whatever a system finds first, the useful next step is simply to see your own verdict. M2.0 builds executive-ready, board-legible AI-search reporting, including a white-labelled AI Visibility Check that shows what ChatGPT, Perplexity and Google's AI answers currently say about your brand and which sources are driving it. It is a way to put the actual picture in front of the people who now need to own it.
