The Logo Test: Strip the Logo. Can a Machine Still Tell It's You?

Take everything your brand published last quarter. Remove the logo, the name, the colours, the URL. Hand what is left to a machine that has read your whole category, and ask one question: who wrote this?

That is the logo test, the simplest brand distinctiveness test there is. Most brands fail it.

The short version: An answer engine can only name what it can attribute, and it can only attribute content that is recognisably yours without the branding. The logo test measures exactly that, with two instruments: a blind attribution check and a measured distance from the machine's own default writing. Fail it and no amount of publishing gets you named. Pass it and you own the position the answer engines are handing out: one of the two or three names in the answer.

Why attribution is the whole game

When a buyer asks ChatGPT, Perplexity or Google's AI Mode who is best, the machine reads everything it retrieved, compares, and names two or three brands. It has no loyalty to your logo, because it never sees your logo. It sees language. If your language reads as the category's language, the machine files your content under the category, and the credit for everything you publish disperses into the average. This is why publishing more cannot fix an attribution problem: the machine cannot credit what it cannot recognise.

The concentration is the proof: 3% of brands capture 71% of AI recommendations. The named few are not louder. They are attributable.

The science: a blind test, not an opinion

The logo test runs on two instruments, and neither involves anyone's taste.

1. Blind attribution. Take the category's content, strip every identity mark, and test whether a classifier can assign each piece to its brand. Accuracy near chance means the brands are statistically interchangeable. This is stylometry, the discipline courts and scholars use to attribute disputed authorship, pointed at brands.
2. Distance from the machine default. Generate the safe, middle-of-the-road version of your topic, the one a generic AI produces on purpose, then measure how far your real content sits from it. That distance is your distinctiveness as a number. Close to the default means the pull already has you.

Together the instruments read the layers that decide attribution: Entity, the identity a machine resolves by name; Voice, the language signature no rival shares; Topic Authority, the proof only you could have produced. Both return evidence, not adjectives: the shared phrases, the formula patterns, the exact places your content and your competitors' content collapse into each other.

The convergence they detect is not a style opinion. In Science Advances, Anil Doshi and Oliver Hauser measured it directly: "generative AI–enabled stories are more similar to each other than stories by humans alone." The logo test tells you whether your brand sits inside that similarity or outside it.

The shape of a fail

A brand we audited had published hundreds of articles: templated series, programmatic guides, weekly round-ups. We read 136 of them and the Brand Voice Score came back at 22 out of 100: a real voice existed in places, and templated phrasing dominated the corpus. One sharp explainer proved a real voice existed underneath. It was the outlier, not the norm, and to the machine the brand read as its category, indistinguishable from the same handful of competitors on every question.

Fail signals	Pass signals
Phrasing shared with competitors	Signature phrases no rival uses
Formula patterns ("whether you're...", "we offer...")	Named methods and owned frameworks
Numbers parachuted in without judgment	Proof with a stance attached
Flat, same-length sentences	A rhythm a reader could pick out blind

Why you can't prompt your way past it

The obvious shortcut is to instruct the AI to be more original. It cannot comply. As I wrote in AI Is a Machine for the Average: naming the edge does not move the machine to the edge; ask for "original" and you get the statistical centre of everything labelled original.

Distinctiveness enters through human judgment: positions taken, proof built, a voice exercised. And it has to be held, because the pull never stops and the decay only shows across the body of work, never in any single piece.

Run it on your brand

1. Paste your URL. Five agents read your content and your competitors'.
2. The blind test runs: logo off, who wrote this?
3. You get your Brand Distinctiveness Index, 0-100 against your category, with your weaknesses ranked and the evidence attached: the phrases, the overlap, the gaps.

Run the test. Free, no call required. That is how Distinctiveness Engineering starts at Ivanooo: measure first, then build what the machine can finally tell apart.

FAQ

What is the logo test? Strip every identifying mark from your content and test whether a machine that has read your category can still attribute the content to you. It measures attributability, the property answer engines select on.

Can strong visual branding compensate for generic content? Not with answer engines. Distinctive assets work on humans; the machines deciding recommendations read language. You need both, and this test measures the language half.

How many brands pass? Most fail. The common verdict in our audits is a real voice buried under templated volume: the distinctiveness exists in places, and the body of work averages it away.

Is this the same as AI-content detection? No. Detection asks whether a machine wrote your text. The logo test asks whether your text is distinguishable from your category's, whoever wrote it. A human team can fail it; an AI-assisted team with strong direction can pass it.

What happens if we fail? Failing returns a map, not a verdict: which phrases you share, which proof you lack, which surfaces you are missing. That map is the work plan for getting named.