Positioning That Survives an AI Summary

Positioning has always been hard to audit because everyone grades their own work. There is now a test that grades it for you, and it takes about a minute.

The test

Ask ChatGPT, Gemini and Perplexity to describe your company in one sentence. Then compare what comes back to the positioning statement you believe you have.

What you receive is a compression of everything published about you. If your position survives that compression, it is real. If what comes back is a generic description of your category, it never reached the page.

A summary is positioning with the marketing removed. What is left is what you actually own.
– The Antimony position

Why most brands fail it

The position lives in the deck, not the site

Positioning is agreed internally, expressed in a strategy document, and then diluted into safe category language by the time it reaches public copy. The model can only read what is public.

It was never distinctive to begin with

The test that predicts failure: could three competitors publish the same statement unchanged? If so, a summary of your material and a summary of theirs will converge, correctly.

Nothing corroborates it

A claim appearing only in your own words is weakly held. Positioning corroborated by client evidence and third-party mention survives compression far better – the mechanism is in Third-Party Mentions Now Beat Backlinks.

What survives compression

Specificity survives. Named clients, real figures, a stated refusal, an unusual constraint. Blanc Bridal as slow, intentional couture in a market competing on trend and price is compressible without becoming generic – the refusal is the position and it is stated plainly.

Adjectives do not survive. Innovative, trusted, end-to-end and premium compress to nothing because every competitor supplies the same words, so the summariser drops them as uninformative.

In practice: what our own summary revealed

Our findability baseline scored GEO 61, and the diagnosis was not that we lacked a position – it was that the evidence for it was disconnected from the material stating it. Our case studies carried the specifics, and no article linked to them. Our articles carried no author or date, so the claims were unattributed.

The general rule: your position is only as compressible as its evidence is reachable. Before rewriting a positioning statement, check whether the proof of the current one is linked from the pages that assert it – because a summariser can only use what it can find and attribute.

How to fix a failed test

Write the sentence you want returned. Then find every public page that should support it and check that it does – specifically, with figures and names rather than adjectives. Then link those pages to each other so the claim and its evidence sit within reach of one another.

Then re-run the test in a month. The summary changes as the material does, which makes this one of the few positioning exercises with a feedback loop.

By the numbers

• GEO 61 / AEO 52 – our own baseline before the evidence was connected.
• 0 of 19 articles linking to the case studies carrying the specifics.
• Under 20% overlap between top Google results and AI-cited sources – ranking will not fix a positioning summary.

Key takeaways

• Ask three AI systems to describe you in a sentence and compare it with your intended position.
• Failure usually means the position lives in the deck rather than the public material.
• Specificity, refusals and named evidence survive compression; adjectives do not.
• A position is only as compressible as its evidence is reachable.
• Re-run monthly – the summary moves as the material does.

Appendix & sources

Definitions. Compression: the summarisation a model performs when asked to describe an entity briefly.

1. Citation and overlap data: seo.com.
2. Client results: Blanc Bridal case study, and the Antimony findability baseline, 26 August 2026.