Fix damaging autocomplete & related searches

How suggestions get removed

Google removes predictions that violate its policies — harassment, unverified accusations, sensitive personal information — through a dedicated reporting channel that responds to precise, well-argued filings. Where a suggestion doesn't violate policy, it can still be displaced: suggestions reflect search behaviour and content, and a structured campaign changes what the algorithm has to work with. We file first, displace second, and monitor because suggestion ecosystems shift.

Where damaging suggestions come from

Autocomplete and related searches are a mirror held up to search behaviour. Google's systems predict queries from what people actually type, what they type next, and the content that exists to satisfy those queries. A damaging pairing like your brand plus 'scam' usually traces to one of three origins: a burst of genuine searching triggered by an incident or a viral post; content farms and complaint sites deliberately optimised for those pairings; or, in the ugliest cases, manipulation — coordinated searching and seeded content intended to manufacture the suggestion. The suggestion then becomes self-sustaining, because every user who sees it and clicks it out of curiosity votes for its continued existence.

Diagnosing the origin is the first real step, because it dictates strategy: policy filings reach suggestions that violate Google's rules regardless of origin, while displacement only works when you understand what behaviour and content are feeding the prediction you're trying to starve.

Mistakes that entrench a bad suggestion

The instinctive responses mostly make it worse. Repeatedly searching the offending pairing yourself — or asking colleagues to check whether it's still there — feeds the exact signal that sustains it. Publishing content that repeats the damaging phrase verbatim ('No, [Brand] is not a scam') can strengthen the query-content association it means to fight; there are ways to address the underlying concern without reinforcing the string. And ignoring the underlying results is the strategic error: a suggestion pointing at a page of removable complaint-site content should trigger removal work on those results too, or the cleaned suggestion regrows from the surviving content beneath it.

This is why suggestion work at our end is rarely a single filing. It's a small campaign: report what violates policy, remove or de-index the feeder content where grounds exist, and shape legitimate replacement demand — in that order.

Measuring success and holding it

Success is defined precisely at the start: which suggestions, for which seed queries, on which surfaces — autocomplete, the related-searches block, and the 'people also ask' features that draw on similar systems. We measure from clean sessions across locations, because suggestions vary by geography and personalisation, and a fix that only shows on your own laptop isn't a fix. Completion means the damaging pairings are gone or displaced from the surfaces that matter, documented with dated captures.

Durability depends on the ecosystem that produced the suggestion. A pairing born from a passing incident, once cleared, tends to stay cleared as search interest fades. One fed by active complaint content or a persistent adversary can regrow, which is why suggestion monitoring — routine checks of the pairings that once hurt you — is folded into our Protection Plans for brands and individuals whose names carry commercial weight.