What it examines, how it works and what an organisation receives.
An AI reputation audit is a controlled investigation of how AI systems describe an organisation, which evidence shapes those descriptions and where the resulting picture creates reputational risk or opportunity.
A handful of prompts produces observations. A proper AI reputation audit tests recurring conclusions, traces public evidence and provides a basis for action.
The thing being audited is an organisation's AI reputation. AI reputation (AIR) is the picture of an organisation that people get when they ask AI about it, and it is built from what the organisation says about itself, what it appears to be doing and what others have said about it.
A journalist researching a company, an investor considering a business or a potential customer comparing suppliers might begin by asking an AI assistant for a summary. That answer could shape their first impression before they visit the organisation's website or speak to its representatives.
Depending on the product and question, an AI system may draw on information already represented in its model, material retrieved from the web or both. The answer may combine corporate claims with reporting, reviews, recruitment and other material. It can also be incomplete, outdated or wrong.
An AI reputation audit examines the picture being presented and whether the public record supports it.
Adelbury's September 2026 report, The AI Reputation Audit, groups public evidence into three layers. The framework does not mean every AI system consults the same sources.
Declared narrative is what the organisation says about itself. Its website, company profile and leadership biographies communicate its identity, priorities and expertise. An audit considers whether those claims are clear, current and supported by evidence.
Observed behaviour is what the organisation appears to do. Job adverts can indicate where it is investing. Executive commentary may reveal its priorities. Product pages and employee activity offer further signs of how the business operates.
External validation is what others record. News reporting, reviews, specialist commentary and public filings may support the organisation's position, challenge it or provide important context.
The layers can disagree. A job advert can qualify a strategic claim. A product page can contradict an executive position. An old article can supply the frame through which newer evidence is read. None of these shows which version is correct. Each tells the audit where to investigate.
The aim is to find where these layers agree, contradict one another or leave a significant claim without enough support.
An AI reputation audit has five stages.
First, define the decision context. Identify the audiences, markets and questions that matter.
Second, build a controlled evidence set. Test a structured set of questions across major AI platforms, repeating important queries rather than relying on one answer. Record when and where the test was run, which product was used, what it said and which sources it cited.
Third, trace the evidence. Work backwards from recurring conclusions to the public material that may support or contradict them. Classify that evidence against the three layers.
Fourth, diagnose the gaps. A persistent contradiction in the public record is different from an outdated source, a difference in how AI products retrieve information or a mistake generated by the AI system. The response depends on understanding which problem has occurred.
Finally, prioritise the response. Assess findings by their potential reputational effect, the strength of the evidence and the organisation's ability to influence the issue. Assign responsibility to the relevant team and decide what should change. Repeat the assessment as the evidence and systems evolve.
A completed AI reputation audit provides a documented baseline of how the tested AI platforms describe the organisation. It maps recurring conclusions and their sources, assessing where the three layers agree or conflict.
It also produces a ranked set of risks and opportunities, with an action plan identifying priorities, owners and measures for reassessment.
Some actions may involve updating inaccurate information or making credible evidence easier to find. Others may require decisions by HR, product, leadership or legal teams. If the organisation's behaviour is the source of the contradiction, changing the messaging alone will not resolve it.
An AI visibility check asks whether an organisation appears in AI-generated answers, and perhaps how often it is mentioned or cited. Those are useful observations, but they do not establish what impression the answer creates.
An AI reputation audit examines what those appearances add up to. It considers the conclusions a reader could take away, the evidence supporting them and the response they warrant.
An AI reputation audit cannot guarantee particular rankings, citations or descriptions. AI outputs vary by platform, model, question, date and access to current information. Citations do not necessarily reveal every influence behind an answer.
Nor can one prompt establish a reliable pattern. Repeated testing and human judgement are needed to distinguish significant findings from isolated errors.
An organisation may find an AI reputation audit useful when its reputation affects decisions by customers, investors, employees or other stakeholders, particularly if its strategy, leadership or offer has changed and the public record may not yet reflect that change.
The report The AI Reputation Audit explains the three-layer framework and the methodology in greater detail, including the full source matrix.
Adelbury runs AI reputation audits for organisations. See what the audit includes.
Talk to Adelbury
