01What happens.
Different AI tools reach different parts of an organisation’s published material and answer differently. The organisation’s own current page can be correct throughout. Which answer a person receives depends on the tool they chose.
02Why it matters.
If you check what AI says about your organisation by asking one tool, you have checked one tool. The answer that reaches an employee, a resident or an applicant depends on the tool they chose, and you have no say in that choice.
When the answers differ, some of the wrong ones come back to you as challenges and some do not, and the right ones arrive looking like any other decision. You can count the claims. You cannot tell which of them an AI answer shaped, so the challenges you handle are the part of the effect you can see.
03The case.
Employees across a public body needed the current mileage repayment rate for a claim. The rate is published, and the body’s own page stated it correctly. The employees asked AI, as people now do before they search a website, and different tools returned different rates, one of them wrong.
Claims went in at the wrong rate. Finance queried them, rejected them and asked for resubmission, and each rejection produced correspondence: the employee asking why, finance explaining the rate, the employee pointing to the answer they had received. The cost of handling each claim was greater than the difference in the rate. Employees stopped trusting the answer, and the finance staff stopped trusting the source they were being shown.
The body counted the rejected claims and the hours spent on them. It could not tell which of the correct claims had been shaped by an AI answer, or how many people had been told the wrong rate and had not claimed at all. The claims were visible; the influence on them was not.
The same difference was seen in a university, where applicants comparing 2 tools received different courses, different descriptions and different dates for the same institution. That instance is carried in use case 35.
AAAnow observed one contributor to such differences in an outside-in assessment. On a single search results page, the AI-generated answer at the top and the organisation’s own current page directly below it presented conflicting information about the same organisation. Conventional search had reached the current page through one route, while the AI-related route had not received equivalent access during the assessment, so older and indirect material had a greater opportunity to shape the AI answer shown first. That is one contributor, observed in one assessment; access, source selection and processing all differ between tools, and no single cause is asserted for any case.
The body’s own page was correct, and checking it told the body nothing. What differed was what each tool could reach and chose to read, and no view of that existed inside the organisation.
04Exposure.
The body is held to what it published, on whichever page each tool read, and the page each tool read was the body’s page. The rejected claims were the visible cost, paid in finance time.
The larger cost was confidence: employees who stopped trusting the answer, staff who stopped trusting the source they were asked to apply, and a published rate that no longer settled the question it was published to settle. The same unreliability is documented across government information: the Open Data Institute tested 11 AI models on more than 22,000 questions against GOV.UK content and found they did not reliably prioritise official sources, with some answers significantly wrong (ODI, Feb. 2026).
05What AI Readiness changes.
What the body could not see from the inside was what each AI route received when it approached the estate. AAAnow tests the routes AI uses from the outside, 15 of them, and evidences whether each is served, challenged or turned away, alongside where older material is still reachable and where versions conflict. With that evidence the body can decide which restrictions it wants, record each one, and establish one current position for the material AI is asked about.
Your team makes the changes; AAAnow provides the assessment, the evidence, the guidance and the review.
06Related use cases.
Residents get different answers to the same question, and the cost lands on the council
Contradiction across an estate, where the answers differ by source rather than by tool.
£350,000 to be seen by systems the estate was closed to
Spend on visibility while AI routes were turned away at the door.
Applicants acted on entry requirements the university had already changed
The university instance of the same difference between tools.
07References.
- Open Data Institute, Feb. 2026. theodi.org/news-and-events/news/new-research-questions-the-trustworthiness-of-ai-chatbots-for-government-informationtheodi.orgWe do not enable external links - please copy and paste, you are then certain as to the link being opened
- Lawrence Shaw, Digital Suicide, LinkedIn, 3 Aug. 2026: the outside-in assessment in which the AI answer and the organic result conflicted on one results page.
- The mileage repayment and university instances are AAAnow observed and anonymised.
Next step
Talk it through, in your terms.
The briefing sets this case in your organisation’s terms, and the discussion follows your questions.
Observed, anonymised. No organisation is identified.