Your brand is now in comparisons you’ll never see, being judged by reviews you didn’t know existed

Something quietly shifted in Q1, and most UK businesses haven't clocked it yet. AI assistants have started pulling negative reviews, Reddit complaints, and forum gripes into product comparisons that nobody asked them to do. A prospect asks ChatGPT which CRM to pick. The model decides, on its own, that a balanced answer needs to mention the time your support team had a bad week in 2023 and somebody vented about it on r/sales.

You weren't searched for. You weren't compared. You were conscripted into someone else's evaluation and judged by evidence you didn't know was on file.

This is not the same problem as traditional reputation management, and treating it like the same problem is going to leave a lot of businesses exposed. The Search Engine Journal piece this morning frames it as a sponsored post for a reputation cleanup service, which is fine, but the actual development underneath the marketing is worth taking seriously on its own terms.

The shift: comparisons became audits

For twenty years, reputation management worked on a predictable model. Someone searches "[your brand] reviews" or "[your brand] scam." You try to make sure the first page of results doesn't ruin you. The trigger was always a deliberate query about your reputation.

That trigger is gone. Or rather, it's been joined by a much larger one you can't see.

When someone asks an AI assistant "which CRM should I choose," the model treats that as a synthesis task. It pulls features, it pulls pricing, and — this is the new bit — it pulls sentiment. It scrapes Reddit threads, complaint sites, G2 negative reviews, forum posts, and weaves them into the answer as "context." The user thinks they're getting a balanced comparison. What they're actually getting is a reputation audit you weren't told was happening, conducted on evidence selected by a system optimising for "helpfulness," not fairness.

The user never searched for problems. The model decided problems were relevant.

Why this isn't just SEO with extra steps

The temptation is to file this under "manage your online reviews" and move on. That misses the structural change.

The reputation surface stopped being a place and became a behaviour.

In classic SEO, your reputation surface was the SERP for branded queries. Finite. Inspectable. You could see what was there, and you could work on it. In AI-mediated comparisons, your reputation surface is everywhere a model might decide to look when someone asks about your category. That surface is functionally infinite, invisible to you, and re-derived every time someone runs a query.

You can't audit a SERP that doesn't exist as a fixed object. Two prospects asking ChatGPT the same question on the same day might get answers that mention your brand differently — or one might surface a six-year-old Trustpilot complaint while the other doesn't. The model isn't lying in either case. It's just that "what AI says about your brand in comparisons" is now a probability distribution, not a page.

The reputation surface stopped being a place and became a behaviour.

The four signals that get pulled in

From the patterns now visible in AI-generated comparisons, the negative content most likely to surface shares predictable traits. Recency and volume. Specificity (vague gripes get filtered; detailed complaints with product names and outcomes get weighted as useful). Platform authority (Reddit, Trustpilot, G2, industry forums get treated as trusted). Recurrence across sources (the same issue mentioned in three places becomes a "verified pattern" in the model's reasoning).

If you've been doing this work for any length of time, you'll notice something: those are the same signals that determine whether content gets cited positively in AI answers. The system doesn't have separate machinery for praise and complaint. It has one citation engine, applied symmetrically.

Which means the playbook is also symmetrical. The brands that get cited well in AI-mediated discovery are the ones with recent, specific, multi-platform, recurring positive signal. The brands that get punished are the ones with the same shape of negative signal and not enough positive signal to balance it. There is no separate "AI reputation strategy." There is the strategy of having a healthy footprint on the platforms AI systems trust, full stop.

What this means if you're a UK SMB

Most of the businesses I work with don't have an active reputation problem in the traditional sense. They have something more dangerous: a sparse footprint. Three Google reviews from 2022. A Trustpilot page nobody's claimed. A Reddit mention that's actually a competitor in disguise. Nothing actively bad. Nothing actively good either.

In the old model, this was fine. You ranked for your brand name, your site looked professional, and reputation only became a topic if something went wrong.

In the new model, sparse footprint is the problem. When a model is synthesising a comparison and reaches for context about you, it grabs whatever it can find. If the only specific, recent, detailed thing it can find is one annoyed Reddit post, that post becomes the representation of your brand in that answer. Not because it's representative — because it's the only signal with enough density to register.

The fix isn't reputation management in the traditional sense. It's making sure the signal-rich content about you skews positive, recent, and specific enough to actually surface. That means the same fundamentals that drive AI citation generally — earned mentions on platforms models trust, customers writing detailed reviews on Trustpilot and Google rather than just leaving stars, presence in the Reddit and forum conversations where your buyers actually hang out.

It is, depressingly for the people selling AI reputation tools, mostly the same advice anyone competent would have given you about brand-building in 2018. Just with more urgency, because the cost of a sparse footprint went up.

The honest part nobody's writing about

Here's the bit the sponsored articles skip. You cannot fully control this. The model will sometimes surface things you'd rather it didn't, regardless of how much positive signal you build. Models hallucinate. They misquote. Fast Company has reported documented cases of AI engines misrepresenting brand statements in comparisons — not pulling real negative content, but inventing or distorting it.

There is no defensible position from which a small business can audit every AI-generated comparison they appear in. The query space is too large. The outputs are non-deterministic. The platforms don't give you logs.

What you can do is two things. Run the comparison queries yourself, periodically, on the platforms that matter — ChatGPT, Perplexity, Gemini, Google AI Mode — for your brand against your top three competitors, and see what the models say. Screenshot the answers. Note what's wrong, what's missing, what's surprisingly accurate. This is the closest thing to a SERP audit available, and it takes maybe twenty minutes a quarter.

Then build the positive signal density that gives the model better material to pull from. Not as a defensive crouch. As the actual job of being a credible business in 2026.

The close

The autonomous reputation surfacing thing is real, and it's going to catch a lot of UK businesses off guard over the next twelve months. Not because it's catastrophic — most won't have a major issue. But because the assumption that "we don't have a reputation problem" stops being valid when the model gets to decide what counts as a reputation problem.

The brands that come through this well aren't the ones who hire reputation cleanup services. They're the ones who already had enough genuine signal — real reviews, real Reddit presence, real earned coverage — that the models had good material to work with. Everyone else is going to discover, slowly and uncomfortably, that being invisible is its own kind of vulnerability now.