Home / Insights

When AI gets your product wrong, and how to fix it at the source

By Abhiraj Aggarwal, founder. Updated 28 September 2026.

The angriest email I have ever received from a prospective client was three lines long. It contained a screenshot of an assistant telling a shopper that her skincare range contained fragrance. She had reformulated to remove all fragrance two years earlier. For her customer, people with reactive skin, that one sentence ruled the product out entirely.

Her question was reasonable. How do I make it stop.

The answer is not satisfying at first and then becomes very practical. You cannot argue with the assistant. You can only fix or outdate the sources it read.

Wrong facts have addresses

An assistant did not invent the claim. It read it somewhere. In our experience a wrong answer is almost always one of four things.

An outdated third-party listing. A retailer, distributor or marketplace page still describing an old formula, an old price, an old size range. These are the most common and the most fixable, because someone on the other end has a duty to keep them accurate.

An old community thread. A forum or Reddit discussion from before you changed something. Nobody is going to edit a three year old post. This is the category that requires patience.

Your own site contradicting itself. Two pages describing the same product differently, or a spec in an image that disagrees with the text beside it. Assistants faced with contradictory evidence from the same brand tend to hedge, or pick one at random.

An inference from the above. Once an assistant believes two wrong things, it will derive a third. "Contains fragrance" plus "contains essential oils" becomes "not suitable for sensitive skin", which is a sentence nobody ever wrote about you.

Trace it backwards

Before fixing anything, find out where each claim lives.

Ask the assistant the question that produced the wrong answer, then ask it directly what sources it used. Follow every link. On assistants that show citations by default, this takes minutes. On the ones that do not, ask for the sources explicitly and treat what comes back as a lead rather than gospel, because it is reconstructed after the fact and is not always complete.

Then search the exact wrong phrase. If an assistant says your product is only sold online, search for that claim as a sentence. You will usually land on the page that started it.

Do this for every wrong fact, not just the one that made you angry. There are normally between three and six, and they are connected.

Keep a ledger

One table, four columns, and it becomes the single most useful document in the project.

The claim. Where it came from. Who can change it. Status.

For example: "Products contain fragrance", outdated retailer listing, retailer's content team, fixed at source. Or: "Products contain essential oils", forum threads about the pre-2024 formula, nobody, outdated by newer sources.

The ledger matters because this work is slow and diffuse, and without it you lose track of what you have already chased. It is also the thing that shows a founder, at the end of the quarter, exactly what changed and why.

Three possible outcomes

Every wrong fact ends in one of three states.

Fixed at source. You contacted the retailer, the directory, the marketplace, and they corrected it. Always try this first and always be specific in the request. A polite email naming the exact page, the exact sentence and the correct replacement text gets actioned. A general complaint does not.

Outdated by newer and better sources. You cannot edit a 2023 thread. What you can do is make sure that the most recent, most credible material about the product says the right thing loudly. Current reviews, current coverage, a clearly dated page on your own site stating the change and when it happened. Over time newer evidence outweighs older evidence.

Unfixable, and therefore your job to answer. Some claims will persist. The response is a page on your own site that addresses the question head on, in the words buyers use. If the confusion is between two terms you have used interchangeably, pick one and use only that one, everywhere, forever.

How long this takes

Weeks, not days. Retailer corrections can land inside a fortnight. Community threads losing their weight takes months and depends on what else gets published in the meantime.

The thing that speeds it up most is ruthless consistency on your own properties, because that is the only part you fully control and it is the evidence that is most obviously about you.

What does not work

The feedback buttons inside the assistants. We have submitted corrections through them at volume, on multiple platforms, and have never been able to attribute a single answer change to one. They may be useful to the companies building the models. They are not a fix for your problem.

What changes answers is changing what the answer is built from.

Where to start

  1. Ask your five most important buyer questions and write down every factual error, however small.
  2. Trace each one to a page.
  3. Build the ledger. Four columns is enough.
  4. Send three correction emails this week, to the three most-cited wrong sources.
  5. Fix any contradiction on your own site the same day you find it.

See where your brand stands in AI answers

Get a free AI visibility snapshot. We test real buyer questions across the major AI assistants and send you where you rank, within 24 hours.