Dil Phagura, Co-Founder and CEO of AI Reputation
An executive called me three weeks before a close. He had not found a new article, and nothing on page one of his name had changed in a year. He had asked an AI assistant to describe him the way a stranger might, and the paragraph that came back was fluent, organized and wrong in one sentence near the middle.
The sentence was not defamatory and it named nothing that could be sued over. It attached him to a category of coverage he had spent years on the right side of, and it did it in the same flat, confident voice these systems use for the weather. What bothered him was not the claim. It was that he had no idea how long the answer had been saying that, or how many people had already read it.
Why what AI says about you is different from what Google shows you
His name query was clean. Two owned profiles, a firm biography, a conference listing, a professional directory entry, nothing adverse in the first ten results. If the only check he had ever run was the one almost everybody runs, he would have closed the round without knowing the problem existed.
Ranking and summarizing are different operations and they reward different things. A ranked list is a competition for a small number of visible positions, so a weak source loses and disappears below the fold. An answer engine is not running a competition. It reads across a body of material, weighs what it finds, and compresses the result into a paragraph that carries no visible hierarchy at all.
That difference has a consequence people underestimate. A source too weak to hold the tenth position in a ranked list can still supply a clause in a summary, because the summary is not ranking it against anything. The reader sees one paragraph with no indication of which part came from a national publication and which part came from a forum post in 2019.
This is the part of my own field I find people resist most, including some of my colleagues. A buried result is not a neutralized result any more, and suppression as a complete strategy rested on an assumption about how people find things that stopped being true.
How to check what AI says about you
I ran the same procedure on his name that I run at the start of every engagement. It takes about an hour the first time. The order matters more than the tool list, because the order is what keeps you from reacting to the first alarming sentence you read.
The queries to run
Start with the bare name. Then the name plus profession, then the name plus current employer, then the name plus former employer if there is one. Then run the query a skeptic would run, which is the name plus the worst word anyone has ever attached to it.
That fourth query is the uncomfortable one and it is the one that earns the exercise. Nobody volunteers to type it about themselves. It is also, in my experience, where the difference between a clean check and a useful check shows up.
Run the whole set through more than one system. ChatGPT, Gemini and Perplexity return meaningfully different portraits of the same person, and Google now returns its own summary above the links, so run that one with AI Overviews visible rather than scrolling past it to the results you are used to reading.
What to record
Record four things for every answer. The exact wording of any adverse clause, the sources cited underneath it, whether those sources are about you at all, and the date you ran it.
Screenshot the answer rather than paraphrasing it. A paraphrase written by the person being described is never neutral, and in three months you will want to compare wording rather than impressions.
Open every citation before you conclude anything. Under the wrong sentence about my client, two of the three sources were about a different company with a similar name, and establishing that took minutes rather than weeks. It also changed the entire shape of the work that followed.
Then run the full set again two or three days later, signed out, from a different device if you have one. Answers move between sessions for reasons you have no part in, so one pass is a sample and not a verdict.
Finally, set something standing so the next change does not wait for a transaction to reveal it. Google Alerts covers the page side, costs nothing and takes four minutes to configure. The answer engine side still has to be re run by hand on a calendar reminder, which is the unglamorous part of this that nobody sells.
How long it takes to change what an answer engine says
Here is what was actually done for him, in the order it was done. The two miscited sources were raised with the publishers first, because a factual correction at the source is the only move in this field that is both cheap and permanent.
Second, a small set of owned and third party assets was built that stated the correct affiliation in plain, unambiguous language. Not promotional writing. Sentences a retrieval system can lift without having to interpret anything.
Third, the strongest of those assets was placed where the retrieval was already reaching, which meant publishing into the sources the answers were already citing rather than the sources we would have preferred them to cite.
The wrong clause stopped appearing in most sessions at around four months. It was intermittent for roughly two months before that, appearing and disappearing with no pattern we could read, and that intermittent stretch is the part clients find hardest to sit through.
None of that timeline was negotiable by spending more money on it. The assets that move retrieval are the same assets that take months to earn, and anyone quoting you a week is describing something that will not hold or something that will make it worse.
What a single check cannot tell you
A check tells you what one system said to one session on one day. It does not tell you what that system says to a recruiter in another country on a different account, and nobody outside these companies can see the weighting behind any of it, including me.
It also cannot tell you whether the clause is costing you anything. Plenty of them sit there for years and never once surface in a conversation, and treating every adverse sentence as an emergency is how people spend real money solving nothing.
And if what you find turns out to be accurate, the check has not handed you a removal problem. It has handed you a question about what you build next, which is slower work and more honest work. Where the underlying matter is legal, it belongs with a lawyer rather than with me. Where it is a privacy exposure rather than a ranking problem, Google's process for removing private information from Search is a separate track with conditions of its own.
The check is cheap and the not knowing is expensive. Almost nobody I work with found out early, and the ones who did found out because they looked on an ordinary Tuesday with nothing at stake.
Run the check before something forces you to.
Dil Phagura is the Co-Founder and CEO of AI Reputation. Connect on LinkedIn at https://www.linkedin.com/in/dilphagura
