A UK employee wellbeing platform's website tells visitors it is rated 4.7 stars. Google's knowledge panel for the same company shows 1.0 star - from a tiny handful of stray reviews on a business profile nobody at the company had claimed or managed.
Both numbers are real, in the narrow sense that each exists on a live surface. The 4.7 rests on six on-site reviews; the 1.0 on a forgotten profile a couple of disgruntled strangers found before anyone else did. Neither would survive five minutes of scrutiny as a fair measure of the product. But machines assembling an answer about this company do not conduct scrutiny hearings. They read surfaces - and these two surfaces disagree by 3.7 stars.
This case study is about reputation contradictions on surfaces you control but forgot you control, what the answer engines currently do with them, and the competitor SERP detail that turns this from housekeeping into strategy.
6 hand-picked reviews
unclaimed profile
§ An unclaimed profile is still your profile
The uncomfortable taxonomy first. The Google Business panel feels like a third-party surface - it lives on Google, you never made it, you cannot edit reviews. But in every sense that matters it is a controlled surface: claimable by the company, manageable by the company, and read by machines as a statement about the company. Unclaimed does not mean neutral. It means the statement is being written by whoever bothers to show up, which - on a profile with no owner - is disproportionately the angriest people in the queue.
Meanwhile the on-site 4.7 has the opposite defect: a rating computed from six hand-picked reviews, presented with the same visual authority as a rating built on thousands. Machines can count the six. A 4.7 from six reviews next to a 1.0 on the company's own Google panel does not read as "great product, small sample." It reads as a contradiction between what the company says about itself and what its public record shows.
§ What the engines currently do with it - and why "currently" is the operative word
Here is the measured part, and it cuts both ways. In my August 2026 spot checks, when ChatGPT was asked directly about reviews of this company, the 1.0-star panel did not surface. The engine answered that it could not find substantial independent reviews, quoted the company's own satisfaction figures, and explicitly caveated them as vendor-supplied. No leak - this time.
Read that answer again, though, because it is not a reprieve. "No substantial independent reviews" is the engine describing a vacuum, and vacuums are the precondition for every sourcing disaster in this series. The next model refresh, the next retrieval variant, the next user asking slightly differently - any of them can pull whatever fills the vacuum, and right now the most prominent machine-readable rating attached to this brand on the open web is the 1.0. The contradiction has not cost an answer yet. It is sitting there, indexed and current, waiting to.
And one aisle over, the same mechanism is already paying a competitor. In the live UK SERPs for this category, a rival ranks with a 4.8-star rich snippet built on 1,291 votes - review structured data, at scale, showing up as stars in the results themselves. The review layer is not a hypothetical machine surface. It is visibly deciding what searchers see on this exact SERP, in this exact category, today.
§ The fix is a decision, then a grind
The technical work here is almost trivial: claim the profile, respond to the reviews, and decide what an honest on-site rating claim looks like while the independent review base is thin. The real fix is the part with no shortcut - actually accumulating independent reviews at a volume that makes the honest number robust, on the platforms machines already trust.
What I put to the client was a sequencing argument: this belongs before, not after, the content push. Every citation win makes more machines assemble more answers about the company - and answers get assembled from whatever surfaces exist. Winning AI visibility while your most prominent public rating is a 1.0 star is building an audience for the contradiction. Fix the record first, then invite the machines to read it.
§ The audit anyone can run before lunch
- Search your company name and read the knowledge panel as a machine would: rating, review count, category, address. Then check whether anyone in the building actually claims and manages that profile.
- Inventory every rating that appears on your own site, and note its sample size. Any rating you publish without a visible, defensible basis is a contradiction waiting for a counterparty.
- List your review surfaces - Google, the software marketplaces, the employer-review sites - and record the rating and count on each. The spread across them is what a machine sees when it goes looking.
- Ask the engines about your reviews directly, multiple runs. Today's answer is your baseline; the point of the baseline is noticing when the sourcing shifts.
- Check whether competitors in your category carry review rich snippets in the live SERPs. If they do, the review layer is already a ranking surface in your market, and abstaining is a position too - just not a neutral one.
Findings dated as of the May and August 2026 audits: panel and on-site ratings verified by direct observation, engine behavior from August spot-check runs, competitor snippet from live UK SERP pulls. The site is anonymized as a matter of client confidentiality.
The habit this case study argues for: treat every surface a machine can read as a statement you are currently making - including the ones you forgot you owned. This company was making two statements about the same product, 3.7 stars apart. The machines just have not chosen between them yet.