Burying a Negative Result No Longer Buries It

If you have ever been asked to suppress negative search results for a brand name, you have been handed the same plan: publish a lot of positive content, keep publishing it, and the bad result slides to page two where nobody looks. That plan now has a public dataset behind it, and the dataset does not support the plan.

On 30 September, Erase.com published an analysis of 714 suppression campaigns launched between August 2024 and May 2026 through MarTech, tracking 2,461 individual negative results from where they started to whether they ever left page one. It is the largest public accounting of this work we have seen, and it is worth reading closely for a reason the write-up does not advertise: the numbers argue against the way the service is usually sold.

Two findings carry the piece. The first is that production volume is nearly irrelevant to the outcome. The second is tucked into the final section, concerns AI Overviews, and quietly removes the finish line that suppression has used for fifteen years.

What 714 campaigns say about how to suppress negative search results

The dataset covers 714 suppression campaigns and 2,461 negative results over roughly 21 months, and its headline numbers describe a starting position far worse than most briefs assume: 43% of campaigns opened with a negative holding position one for the client's own name.

Three quarters had at least one negative in the top three. About 30% started with five or more negatives on page one. The median campaign published 28 to 29 positive assets, roughly ten a month, which is a real content operation by any standard and a real monthly invoice. You can read the full write-up of the 714-campaign analysis for the methodology.

Before going further, name the conflict. Erase.com sells reputation suppression, and this is a report on its own book of business, which means the sample is not a random draw of brands with a bad search result. It is a sample of brands whose situation was bad enough to hire somebody. Expect the severity to run high and the success rates to be presented generously. We would not quote the win rates to a client without that caveat attached.

What makes the data useful anyway is that its central finding is bad for the seller. A vendor with a thumb on the scale would conclude that clients should publish more. This one concludes the opposite, and that is the kind of result worth taking seriously, the same way we weigh any vendor-sourced recommendation against what the sources actually support.

Starting position decided the outcome, not output volume

The single strongest predictor of clearing page one was where the worst negative result started, not how much positive content the campaign produced: campaigns whose worst negative began at position four or lower cleared page one about 3.5 times as often as campaigns fighting a negative at position one.

Campaigns starting from positions two or three cleared about twice as often as those starting from position one. Negative count compounded it: campaigns with a single negative cleared page one roughly four times as often as campaigns carrying six to ten. None of that is surprising on reflection. All of it is ignored by the way these programs get scoped, because scope is written in deliverables per month and deliverables per month is the one variable that turns out not to matter much.

Here is the number that should reset the pricing conversation. Successful campaigns placed a median of six new assets in the top 10. Unsuccessful campaigns placed four. That is the whole gap. Both groups published 28 to 29 assets. The winners did not write more, they landed two more of the same volume into positions that count.

“Both groups published the same amount. The difference between clearing page one and not clearing it was two additional placements inside the top 10.”

So this is a placement problem wearing a production problem's clothing. A program that reports assets published is measuring the input that does not discriminate between success and failure, which is the same category error as reporting content output instead of measuring visibility with honest error bars. The only number worth putting on a suppression dashboard is how many owned or controlled URLs currently hold top-10 positions for the exact name query, tracked weekly.

The asset mix explains part of the shortfall. In the page-one sample, contributed articles and interviews made up about 35%, press releases and syndicated news 28%, owned websites 19%, and business or bio profiles 13%. Press releases and syndication are cheap to produce and weak at holding a position, which is precisely why they are 28% of the output and a small share of the result.

The timeline is the one piece of good news for sellers of this service, and it is genuinely good: 85% cleared by month four and 99% within six months, with a median of four months across the 46 documented cases sampled. The slowest case in that sample ran 16 months. If you are setting expectations with a legal or communications team, four to six months is defensible and anything promising six weeks is not.

The negatives that will not move are the ones nobody budgets for

The study ranks negative result types by how often they left page one, and the ranking inverts most people's intuition: court records, mugshot sites and arrest aggregators were the easiest to clear, while Reddit threads, reviews and complaint profiles were the hardest.

Court records, mugshots and people-search listings often came down within about two weeks, because they are usually removable at source through a request or an opt-out rather than something you have to outrank. Local news, social media, government pages and national outlets sat in the middle. Reddit threads, reviews and complaint profiles sat at the bottom: court records left page one more than twice as often as Reddit threads did.

That Justice Department finding is a clean natural experiment: the pages are identical in template and authority, and the only variable is age. Releases nine or more years old left page one nearly five times as often as those five years or newer. Some of what gets billed as suppression is time passing, and an honest program says so rather than invoicing for the calendar.

The hardest tier deserves its own line item, because the reason Reddit threads will not move is structural. They are not competing with your content on authority, they are hosted on a domain that wins on name queries almost by default and carries engagement signals your bio profile cannot match. Treating that as a ranking fight is a budget mistake. The only durable approach is participation, which is slower, cheaper and governed by rules that punish anything that looks like a plant, as we have covered in detail on how credible Reddit presence is earned rather than bought.

Why page two no longer suppresses negative search results

The study's final section reports that only 37.9% of AI Overview citations came from pages ranking in the top 10, while 31.2% came from positions 11 to 100, which means the page-two burial that defines success in suppression work no longer takes a result off the surface most people now read first.

Sit with what that does to the win condition. The entire discipline is built on a single assumption, stated plainly in most guides to this work, that moving a negative result to page two makes it functionally invisible because almost nobody clicks to page two. That assumption was true about human clicking behaviour and it remains true about human clicking behaviour. It is not true about retrieval. An answer engine assembling a response to a name query is not clicking anything, and a third of what it cited came from beyond the top 10.

A negative result at position fourteen is therefore not buried. It is out of sight of the user and fully in scope for the system writing the summary the user reads instead. Worse, the summary sits above every organic result you spent four months placing, so a campaign can clear page one, report success, and leave the client with a negative claim restated at the top of the page in Google's own voice.

This is the same mechanic we write about constantly from the positive side. Pages that get cited are not always the pages that rank, because retrieval rewards a passage that answers a question cleanly and independently of where its URL sits, which is the whole argument in how answer engines pick what to cite and in our teardown of what an earned AI citation is actually made of. Every reputation program now inherits that mechanic in reverse. If a passage inside a negative article is the cleanest available answer to the question being asked, its ranking position is a weak defence.

The practical consequence is that the positive assets have a second job. They are no longer there only to occupy positions, they are there to be the better source: more specific, more current, more directly responsive to the question, and structured so a single passage can be lifted without the rest of the page. That is the work we do under generative engine optimization, and it is the part of a suppression program that most vendors are not yet staffed to deliver.

What the vendor's own numbers do not settle

Three limits in the dataset matter enough to state before anyone builds a plan on it, and none of them are disclosed as prominently as the headline percentages.

First, the sample is a vendor's client list, so selection bias runs in both directions: severity is higher than the population average, and campaigns that were abandoned early may not be counted the same way as campaigns that ran to term. Of 2,461 negatives tracked, 2,239 were labeled as either pushed off page one or remaining, which leaves a couple of hundred unaccounted for in the framing. Second, success is defined as clearing page one for a name query, and the study itself then supplies the evidence that clearing page one is no longer the same as resolving the problem. Third, the AI Overview citation figures are reported without a stated sample size or query set in the summary, so treat 37.9% as directional rather than as a number to quote to a board.

How to run a suppression program that survives AI answers

A defensible program in late 2026 runs in four stages and reorders the conventional sequence: audit removability, fix the retrieval surface, place a small number of strong assets, then monitor the answer rather than the ranking.

Start by sorting every negative result into the three difficulty tiers and attacking the easiest first. Opt-outs and removal requests on court aggregators, mugshot sites and people-search listings cost almost nothing and resolve in about two weeks, and they shrink the problem before any content gets commissioned. Check the age of federal and news pages while you are there, because some of the inventory is going to decay on its own and should not be funded.

Then decide, honestly, which negatives are not going to move. Reddit threads, reviews and complaint profiles mostly will not. For those the answer is resolution on the platform or acceptance, not a ranking project, and saying so early is the most valuable thing a consultant does in this engagement. Where the negative result reflects something the organisation genuinely did, the structural fix is operational, not editorial, which is the uncomfortable conclusion we reached writing about how site reputation policies get enforced across regions.

Then place assets, with the placement as the deliverable. Six in the top 10 was the winning median, so commission accordingly: a small number of substantial pieces on domains that can actually hold a position, not thirty syndicated releases. Write each one to be liftable as a standalone answer to the question a buyer is really asking about the brand, because that is what determines whether it gets cited by the summary sitting above the results.

Then change what you monitor. Weekly rank tracking on the name query is necessary and no longer sufficient. Log what AI Overviews, ChatGPT and Perplexity actually say in response to the brand name and the obvious follow-ups, and log which URLs they cite, because that is the only way to see a negative result at position fourteen still doing damage. This belongs in the standing report, not in an annual audit, and it is the kind of measurement we build into reporting and analytics engagements by default.

See where you are cited today

A free snapshot audit of your rankings and AI citations before we ever talk.