AI Search Research

Entity ambiguity is what happens when a machine cannot tell confidently who you are, and the usual result is not a penalty ... it is quiet omission. You do not get a warning. You get less. The fix is almost never more content. It is removing every place where a system has to guess.

We can show you this happening, because it happened to us.

The receipt

We operate two sites in the IBM midrange space with deliberately overlapping subject matter. Call them Site A and Site B. They target similar terms on purpose, which is a strategy we still believe in, for reasons we will get to.

Here is what the search data recorded:

• Site A ... August 2026: 16,067 impressions. First 8 days of September: 392.
• Site B ... August 2026: 13,246 impressions. First 8 days of September: 3,835.

Raw monthly numbers are unfair here, since September is only 8 days. So convert to a daily rate. Site A went from roughly 518 impressions a day to 49, a drop of about 91 percent. Site B went from about 427 a day to 479, slightly up.

Two sibling properties, similar topics, same operator, same technical standard, same week. One effectively stopped appearing. The other did not move.

What we assumed first, and why we were wrong

The first assumption was a penalty. It is the reflex when impressions fall off a cliff, and it is usually the wrong one. There was no manual action, no coverage change, no crawl error spike, and no meaningful difference in technical quality between the two properties.

The second assumption was seasonality. That did not survive contact with Site B, which sat in the same category, on the same server, over the same eight days, and held steady.

What is left is the least dramatic explanation and the most likely one: the two sites were competing to represent the same thing, and a search system consolidated around one of them. Not a punishment. A choice. When two properties look like plausible answers to the same query and are not clearly differentiated as separate entities, something has to break the tie, and you do not get a vote.

We want to flag the limit of this evidence honestly. We cannot see inside the ranking system, we have one clean pair, and eight days is a short window. What we have is a strong, well-controlled correlation, not a proven mechanism. We are reporting it because it is the cleanest natural experiment we have run into, not because it settles the question.

Where ambiguity actually comes from

Very little of it is exotic. In practice it accumulates from four ordinary sources.

A shared or near-shared name

Another organization, product, or person with your name, or a name that differs by a word most people drop in conversation. This is the one nobody can fix by writing more content, and the one most likely to be causing the problem.

A description that changes depending on where you look

Your homepage says one thing. Your LinkedIn says another. Your structured data says a third, usually because it was written a year earlier by someone else. Each version is defensible. Together they force a system to pick, and picking costs confidence.

Relationships that are implied rather than stated

You know that your research property and your services property belong to the same organization. A machine knows it only if something says so. A footer link is a hint. A declared relationship in structured data is a fact.

Siblings that are too similar to each other

This is the one our own data caught us on. Building several properties against the same topic is a legitimate strategy for occupying more of a results page. It stops working when the properties are not distinguishable as separate entities, at which point you are no longer occupying two results, you are asking a system to choose between two versions of one.

The distinction that actually matters

Redundancy and ambiguity look identical on a spreadsheet and behave nothing alike.

Redundancy is two clearly distinct entities that both address the same subject from genuinely different positions. A research property publishing measured experiments and a services property explaining an offer can both rank for related terms, because they are not competing to be the same thing.

Ambiguity is two properties that a machine cannot separate. Same framing, same claims, same structure, interchangeable copy. That is not two answers. That is one answer with a duplicate.

We did not abandon the multi-site strategy after the drop. We sharpened the differences. Distinct positioning, distinct primary subject matter, distinct structured data declaring what each property is and how it relates to the other, rather than two near-identical entities quietly bidding against each other.

Clarity is a design decision

Everything that reduces ambiguity comes down to the same principle: never make a machine infer something you could simply state.

• One authoritative description. Write it once. Use it everywhere, verbatim, including in your markup. Small variations for tone are fine on the page. The machine-readable version should not drift.
• Declared relationships. Founder, parent organization, sibling properties, the specific things you are about. Stated in structured data, not implied by navigation.
• Deliberate disambiguation. If you genuinely share a name with something else, address it. A page that explains what you are not is doing real work.
• Differentiated siblings. If you run several properties in one subject area, make sure each one would be described differently by someone who read all of them.

How you would know it worked

Impressions are a lagging and noisy signal, so we watch something earlier: how many distinct AI crawlers fetch a property. Across our portfolio the range is wide, from single digits to 23 separate AI systems on the best-covered property. A site that is clearly defined tends to get read by more independent systems, not just more often by the same one.

That is a proxy rather than a proof, and we say so every time we use it. But it moves earlier than rankings do, and it points in the same direction.

The uncomfortable part

Entity ambiguity is unusual among visibility problems because the work to fix it is unglamorous and mostly consists of deleting variation. No new content. No campaign. Just a slow pass through everywhere your organization is described, making all of it agree.

It is also the one we see fixed least often, because it does not feel like progress while you are doing it. Our own drop cost us most of a site's search visibility in a week. The fix was not writing anything. It was admitting two properties had grown too alike, and making them different on purpose.