01What happens.
Services, districts and arm’s-length bodies each publish their own version of the same guidance, at different times and in different places. AI reads them together, weights one above the others or blends them, and serves the result with confidence and without a caveat. Trusted repetition then establishes the wrong position at scale.
02Why it matters.
If you run a service, the contradictions in your estate are paid for in your contact centre. A resident who acts on the wrong version calls back, complains, or turns up with the wrong form, and an officer resolves it by hand at the counter or on the phone. The same happens the next day with the next resident, because the versions are still online.
Residents see the confusion, and the cost becomes visible only when someone adds it up.
03The case.
This case is illustrative, drawn from patterns observed across local authority estates and from the documented finding in New York that identical questions put to a city chatbot returned different answers at different times.
A county authority changed the charge for a household service. The service team updated its own page the same week. The district councils that promote the service on the county’s behalf did not, and 3 of them kept the old charge on their pages. A PDF leaflet from 2 years earlier stayed on the county’s document store with the old charge and the old telephone number. The arm’s-length body that operates the service carried a third figure on its own site, set at a mid-year change nobody had passed on.
A resident asking AI what the service costs received one of the 3 figures, chosen with confidence and without a caveat. Which figure depended on the tool used and the day. Some residents received a blended answer that matched nothing the authority had ever charged.
Residents booked at the wrong price and disputed the invoice. Others called to check, which cost an officer’s time each time, or complained when the figure they had received was refused. The service team checked and republished the county page, without effect, because the county page was already right and the other versions were the ones being read. The internal check found nothing for the same reason: an officer checking the charge saw the correct figure and closed the query.
The cost accumulated in 4 places: repeat contact, complaints handled, invoices reworked and the remediation of the pages, county, district and arm’s-length, when the cause was finally found. None of it appeared as one figure, because it was absorbed into the running cost of the service, and the contact centre logged the calls as pricing queries rather than as one problem.
Nobody was wrong, because each version had been correct when it was published. The estate as a whole was wrong, and no single team owned the whole of it, so nobody was looking for the contradiction.
04Exposure.
An authority is held to what it publishes, on whichever property it was published and however long ago. Where 3 positions are live, the authority has published 3 positions, and residents are entitled to rely on the one they received.
The contradiction is paid for from the service budget as repeat contact, complaint handling, rework and remediation, absorbed as running cost and never added up as the single figure it is. The larger cost is the resident’s confidence in the authority’s word, which each contradiction reduces. An authority that can show its guidance has one current version, and can evidence how contradictions are found and settled, carries that into any challenge. New York’s business chatbot showed the same inconsistency on the public record, with 10 city staff asking the same question and receiving different answers (The Markup and StateScoop, Jan. 2026).
05What AI Readiness changes.
What the authority could not see from the inside was the contradiction itself, because each team’s own page was correct. AAAnow reads the whole landscape as AI reads it and evidences where positions conflict across services, districts and arm’s-length bodies, ranked by the exposure each carries. With that evidence the authority can arbitrate which position stands and retire the others.
Your team makes the changes; AAAnow provides the assessment, the evidence, the guidance and the review.
06Related use cases.
The chatbot was replaced twice before anyone looked at what it was reading
Contradiction served by the authority’s own system.
Regulator embarrassment at poor document control
The same problem in a regulator’s estate, with legal consequence.
Planning decisions went wrong on a website the city had forgotten
Versions across inherited properties, and the litigation that followed.
07References.
- The Markup and StateScoop, Jan. 2026: New York City chatbot returning different answers to the same question.
- 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
Next step
Talk it through, in your terms.
The briefing sets this case in your organisation’s terms, and the discussion follows your questions.
Illustrative, drawn from observed patterns. No organisation is identified.