What is a retrievable brand?

Also known as
retrievable brand

A retrievable brand is one an AI system can find, identify correctly and describe accurately when somebody asks about its category. Three conditions, and all three have to hold: the system has encountered you, it knows which entity you are, and what it says about you is true.

Three honeycomb cells in a row: a filled centre cell holding a fishing hook, with a file folder and a simple key in the cells either side.

Also known as retrievable brand.

Where the word comes from, and what we have done with it

Retrievability is not our term. It comes from information retrieval research, where Leif Azzopardi and Vishwa Vinay proposed retrievability in 2008 as a measure of how easily a document in a collection can be found by the queries a system is likely to receive [1]. It was a property of documents in an index, not a marketing idea.

Brave New uses it as the name for a method, applied to companies rather than documents. That is an adaptation, and it should be labelled as one rather than presented as established literature. The underlying insight carries over cleanly: in any collection, some items are structurally easy to surface and others are structurally hard, and that property can be measured and improved independently of quality.

The three conditions, in the order they fail

Presence. Has the system encountered you at all. A company that exists only on its own website, with no third party mention, is close to invisible to a model trained on the open web. This is the most common failure and the slowest to fix, because the remedy is other people writing about you.

Identity. Does the system know which entity you are. Companies that share a name with another company, a product, or a common phrase get merged. The system answers confidently about a composite that does not exist. This is the failure that produces the strangest results and the one most likely to go unnoticed.

Accuracy. Is the description right. A system can know exactly who you are and still describe you by what you did five years ago, or by the one service that happens to be best documented on the open web rather than the one that makes your money.

Most work sold as AI visibility addresses the first and ignores the second and third. The order matters, because improving presence while identity is broken just produces more confident wrong answers.

Why it is not the same as ranking

Ranking is comparative and positional. Retrievability is binary and then qualitative: can you be found, and once found, is the account of you correct.

A brand can be highly retrievable and never recommended, because the model knows exactly who you are and judges you a poor fit. That is a positioning problem and no amount of technical work will fix it. A brand can also be recommended while badly described, which is worse than it sounds, because every recommendation carries the error with it.

Separating the two is the point of the concept. They have different causes and different remedies.

What makes a brand retrievable

Consistency, mostly, which is unglamorous.

The same entity name spelled the same way everywhere. A single canonical description that does not drift between the website, the directory listings, the founder’s profiles and the press. third party sources that corroborate what you say about yourself, because a model weighs independent mention differently from self-description. Structured data that states plainly what the company is. And enough published material in your own category that there is something to retrieve.

None of that is a trick. It is the machine-readable version of being a coherent company, which is why brands with confused positioning tend to be badly described by AI as well as by people.

What to do with it

  1. Test all three conditions separately. Ask an engine who you are, what you do, and who it would recommend in your category. Three different questions.
  2. If identity is confused, fix that before anything else. Disambiguate the name, the location and the category in every place you control.
  3. Write one canonical description of the company and use it verbatim everywhere. Not a version per channel.
  4. Count your third party mentions. If every source about you is published by you, presence is the constraint.
  5. Re-test quarterly and keep the dates. Answers drift, and an undated test tells you nothing about direction.

If you run the company rather than the marketing

The failure that should worry you is not absence. It is confident misdescription, because it is invisible from the inside and it compounds.

If an assistant has merged you with a similarly named company, or is describing a service you stopped offering, it is telling that to every person who asks, identically, with no indication that it is unsure. You will not see it in any dashboard. Nobody will email to tell you. The first signal is usually a prospect arriving with a wrong idea of what you do, and that gets filed as a bad lead rather than as a symptom.

Ask three questions yourself, by name, in one sitting. Who is [your company]. What does [your company] do. Who would you recommend for [your category].

If the first two come back wrong, that is a correction project and it is urgent. If they come back right and the third does not name you, that is a positioning problem and no technical fix will touch it.

Sources

  1. Azzopardi, L. and Vinay, V. (2008). Retrievability: an evaluation measure for higher order information access tasks. Proceedings of the 17th ACM Conference on Information and Knowledge Management (CIKM).

Every source cited across the Hive is listed in the Library.

Get the next one in your inbox

Anne Becheru, who writes the Hive, sends the Brave New Journal every Tuesday: one thing happening in the world, turned into a marketing decision you can make that week. Free, and you can leave from any issue.

Read past issues · Privacy