AI accuracy and brand information

Responding to inaccurate AI information about your brand

Use this evidence-led correction workflow to capture an inaccurate AI answer, verify the approved fact, repair controllable sources and record an honest retest.

Published by SALIENS · Published 2026-09-30

Treat an inaccurate answer as an evidence problem

An AI answer may confuse a trading name with a legal entity, repeat an old service or price, invent a location, omit a qualification, or merge two organisations. Do not begin by rewriting every page. First preserve what was actually shown and decide which statement is inaccurate against an approved source.

Save the exact question, answer, cited links, platform or interface, date and time, market, language, account or session condition, and a screenshot where permitted. AI answers can vary between prompts, systems and sessions. Without this incident record, a later difference cannot reliably be described as a correction.

Complete an eight-part correction record

Use one record for each distinct inaccurate statement. Separate evidence you control from observations of an external system, and keep unresolved items open rather than declaring success after a page edit.

AI information correction and acceptance record
Field	What to record	Acceptance evidence
1. Incident capture	Exact prompt, answer, cited sources, platform, interface, market, language, date, session condition and screenshot.	Another reviewer can identify what was observed without relying on memory.
2. Error classification	Mark the issue as identity, service scope, price, location, contact, credential, dated information or unsupported claim.	The disputed sentence and its business risk are explicit.
3. Approved fact	Write the correct fact, its qualifications, approving owner and controlling source.	The source is public where appropriate, current and approved by the organisation.
4. Source map	List the owned page, localized versions, structured data, PDFs, profiles, directories and third-party pages that repeat the fact.	Each source has an owner, status and action decision; no source is assumed to be causal without evidence.
5. Controlled changes	Correct visible copy, linked evidence, metadata and applicable structured data on sources the organisation controls.	Live URLs show consistent information; structured data matches visible text.
6. External requests	For sources outside your control, record the documented correction route, evidence supplied, submission date and response.	The request has a reference or dated copy; submission is not recorded as acceptance.
7. Release checks	Check status code, canonical URL, language alternatives, internal links, rendered text and relevant schema after publishing.	A dated release record identifies the changed URLs, checks, exceptions and owner.
8. Retest and closure	Repeat the saved question under the stated conditions and label the result corrected, unchanged, mixed or unavailable.	The observed result and date are retained separately from delivery evidence; unresolved variants remain open.

Repair the authoritative source before chasing repetitions

Start with the page that should control the fact: for example, the organisation page for legal identity and contact information, a service page for scope, or an official pricing page for current fees. Make the answer easy for a person to find in visible text, link it from relevant pages and remove contradictions across language versions, downloadable files and official profiles.

Google recommends keeping important content in textual form and ensuring structured data matches visible page text. Its Organisation guidance says structured data on the home page or a single organisation page can help Google understand and disambiguate administrative details, but markup should describe information that genuinely applies. There is no special AI schema or machine-readable file that guarantees an AI feature will use the correction.

Use substantiation and qualifications, not stronger replacement claims

A correction is not permission to replace one unsupported statement with a more favourable one. For every objective service, performance, credential, price or availability claim, retain its documentary evidence and material qualifications. The CAP Code requires substantiation for objective claims and says marketing must not omit significant limitations or exaggerate capability.

If the organisation has changed its offer, keep the effective date and explain what remains available. If evidence is private or conditional, publish only what can be supported and disclose the relevant limitation. Do not create reviews, awards, clients or endorsements to make a correction look more convincing.

Set a realistic supplier scope and closure rule

Before appointing a GEO supplier, ask who approves facts; which websites, profiles and data feeds the supplier may change; which third-party correction routes are included; which markets, languages and AI systems will be observed; how many retests are included; and what evidence closes or escalates an issue. Require the supplier to distinguish completed delivery, source-owner acceptance and an external AI observation.

No supplier controls when a search or AI system recrawls, reprocesses or displays changed information. Google states that meeting requirements and best practices does not guarantee crawling, indexing or serving. A defensible closure rule therefore records the source corrections and repeated observations without promising that every answer will change by a fixed date.

How SALIENS can support an accuracy programme

SALIENS helps brands become discoverable and accurately represented in search and AI-generated answers. We provide GEO audits, AI visibility monitoring, content and technical improvements, and ongoing GEO delivery for UK and international organisations, including scoped English and Simplified Chinese work.

Email info@oxfordintelligence.co.uk with your website, the inaccurate statement, the saved question and answer, the approved source, target market and language, and the systems you want observed. SALIENS can map conflicting sources, define controlled changes and create a repeatable correction record. SALIENS is operated by Oxford Intelligence Limited; source corrections do not guarantee crawling, indexing, citations, rankings, visits or enquiries.

Sources

Prepared with AI assistance using the linked sources and SALIENS service information.