The first hour after someone discovers a negative search result about themselves or their company is usually the worst hour to make decisions. The instinct is to attack the problem: call a lawyer, demand a takedown, buy reviews, blast out five press releases. Almost every one of those reflexes makes the situation last longer than it needed to.
A bad result is an information problem before it’s a legal or PR problem. And information problems respond to the same thing every time: better, clearer, better-sourced information in the places where people — and now machines — go looking for answers.
Why one result feels bigger than it is (and sometimes is)
Over a decade in reputation and search has taught me that the damage from a negative result is rarely proportional to its content. It’s proportional to how much else exists. If someone searches your name and finds a complaint post surrounded by your site, your speaking history, legitimate press, a strong LinkedIn profile and a few industry mentions, the complaint reads as one data point among many. If that same post is the third result on an otherwise empty page one, it reads as the story.
That’s the whole game. Online reputation management is not the art of making things disappear. It’s the work of making sure a reader — a client, a recruiter, a judge, an investor — has enough context to form an accurate picture.
There’s a newer wrinkle, too. When someone asks ChatGPT, Perplexity or Google’s AI Overviews about you, they aren’t scanning ten blue links and weighing them. They’re getting a synthesized paragraph built from a handful of sources the model decided were worth citing. A thin search footprint with one loud negative page doesn’t just rank badly anymore — it feeds a summary. That’s why I keep telling clients that in AI search, citations beat rankings. You can be on page one and still be invisible in the answer.
How search and AI actually decide what to show
Understanding the mechanism keeps you from wasting money. Classic search ranks documents per query. That means a negative result isn’t “out there” in general — it’s specific to certain searches. “[Your name]” and “[your name] reviews” and “[company] lawsuit” are three different battlefields with three different competitive landscapes.
AI systems work differently. They retrieve a set of candidate sources, then generate an answer from the ones that are easiest to extract clean claims from. Clarity and structure make content citable. A well-organized page with plain statements, dates, named entities and clear headings gets pulled into answers more often than a beautifully written essay that buries its facts. This is the core of Generative Engine Optimization (GEO) — optimizing so that AI systems can find, trust and quote you, not just so a crawler can index you.
Here’s the concrete version. Say a founder has an old article about a failed venture ranking for their name. Filing takedown requests probably fails — it’s journalism. Buying a wall of press releases probably fails too, because nobody cites press releases and AI systems discount them. What tends to work is building genuinely useful, well-structured material the founder can stand behind: a substantive site with a clear biography and a body of work, a couple of earned mentions in outlets that actually get quoted, and honest, specific commentary about what they learned. Six months later the old article still exists. It’s just no longer the only thing that exists — and the AI summary now has better material to work with.
What to do, in order
This is the sequence I use in client work, and the order matters more than any individual tactic.
1. Map it before you touch it
Write down the actual queries where the result appears: your name, name plus city, name plus profession, company name, company plus “reviews” or “complaints.” Then run the same questions through ChatGPT, Perplexity and Google AI Overviews and note which sources get cited. You cannot fix what you haven’t defined. Half the time, the “disaster” only surfaces on one long-tail query almost nobody searches.
2. Decide whether this is a removal case or a context case
Legitimate removal paths do exist, and they’re worth pursuing when they apply:
- Factual errors — a correction request to the publisher, with documentation, politely and once.
- Policy violations — doxxing, non-consensual images, leaked personal data and fake reviews can often be reported through the platform’s own process.
- Outdated content — expunged records, resolved disputes and dead pages can sometimes be de-indexed or updated.
- Impersonation — fake profiles and cloned sites are usually the fastest wins available.
What doesn’t belong on that list: abusing copyright claims, mass-filing bogus complaints, or paying a vendor who promises deletion of legitimate journalism. Those tactics get discovered, and getting caught suppressing coverage is a far worse result than the coverage itself.
3. Choose your response posture deliberately
Not every negative result deserves a reply. A public response amplifies the original — it creates a new page, new links and new material for AI systems to summarize. Respond when you have something substantive to add: a correction, an accountability statement, a policy change. Stay quiet when the only available message is “we disagree.” The way I think about this: if your response wouldn’t read well to a neutral stranger in two years, don’t publish it.
4. Build the assets that actually hold
This is where most of the durable progress happens. A proper owned footprint — your site, a clear about page, a body of published work, consistent profiles, structured data — is the foundation of everything else. I’ve written about this in more depth in owning your page one, and about the broader discipline in what online reputation management really is. One thing that consistently works: publish the things only you can publish. Your process, your data, your hard-won lessons. That material earns links and citations on its own merits, which is the only kind of authority that compounds.
5. Earn coverage instead of buying it
Reputation is earned, not bought. Contributed commentary, real interviews, podcast appearances, industry publications, well-argued answers in communities where your buyers actually hang out — these get cited by AI systems because they carry genuine editorial signals. A wall of paid placements on sites nobody reads does not. If you’re deciding where to invest, invest in the sources that get quoted.
6. Make your good material machine-readable
Once the right information exists, make it easy to extract. Clear headings that match real questions. Direct answers in the first sentence of a section. Dates, locations, roles and credentials stated plainly rather than implied. Consistent entity naming across every profile. This is the practical heart of the ARC Method and AI-citation frameworks I’ve built and refined through client work — structure so the machine can quote you accurately, and it usually will. If you’re managing this across a whole company, AI visibility and fulfillment automation turns it from a one-off cleanup into a repeatable system, because systems and automation are what scale quality.
The durable principle
What I’m seeing across AI search is that reputation is increasingly decided before anyone clicks anything. The answer gets formed in a generated paragraph, from a short list of sources. So the strategy that survives is the boring one: show up where the answers are formed, with material that’s accurate, structured and genuinely yours.
That approach is what informs how I work as Head of Fulfillment at Reputation Pros, and it’s the spine of the Cory Maki reputation management philosophy — treat a negative result as a gap in the record rather than an enemy to defeat. Fill the gap with something true and useful, make it easy to cite, and give it time. You won’t erase your history. You’ll just stop letting one page tell it for you.
If you want to manage your online reputation properly, start with the audit, not the panic. Map the queries, check what the AI engines are citing, fix what’s legitimately fixable, then spend the next year building the record you’d want a stranger to find. That’s slower than a takedown request. It’s also the only version that lasts.
Photo by Firmbee.com on Unsplash