Positioning audit, run by an AI in minutes

This page is for anyone who owns a website and wants to know what it actually says to someone arriving cold. It is one of the ready-made studies in the use cases for visual product feedback hub, alongside the pricing page teardown.

What a positioning audit is

A positioning audit is a structured read of your public pages from the point of view of a visitor with no context, deciding in seconds whether this thing is for them. It judges what the site communicates, not what the product does. The question is never "is this feature good." It is "did the page make me understand it, believe it, and know whether it's for me."

A page can describe a great product and still fail all three. The most common way is the hero that positions no one: "for teams of any size, solo or enterprise" names everyone and therefore selects nobody. The second most common is the swap test failure: a headline where a competitor's name could replace yours and nothing else would need to change.

How the AI runs it

You give it your homepage. It opens a review and works through the pages in a fixed order, taking a screenshot of every point it judges and quoting the exact line at issue.

1. The five-second test, first. Before reading anything else, it loads the homepage at desktop width and reacts to the fold alone: what does this do, who is it for, why believe it. It writes those answers down before scrolling, because a first impression is unrepeatable. Then the same three questions from the phone fold.
2. The key-page sweep. Home in full, the product or features page, use-case pages, pricing (read as a messaging surface, not a price check), about, the docs landing, the blog if the nav promotes it. Phone width wherever the layout changes what the message says.
3. Six lenses on every page. Who it's for and not for. Category and claim, and whether the page proves it with specifics or merely asserts it. The swap test. Jargon the target buyer wouldn't know, and load-bearing terms used before they're defined. Consistency between pages. Trust near each ask, and anything actively working against it: dead links, placeholder copy, a blog last touched years ago.
4. The clearest moments. What works, captured with the same rigor, because the do-not-regress list is half the value.

If you run it from your own repository, it also reads your positioning and SEO docs to learn what you intended, checks whether those docs are older than the product work that shipped since, and reports where transmitted and intended diverge.

What comes back

A review at a shareable link. At the top, the positioning statement your site actually transmits, one sentence, phrased the way a cold visitor would repeat it: "it's X for Y that Z." Then findings by severity:

• Garbled. A target visitor would misunderstand what this is or who it's for.
• Muddy. Understandable but unconvincing or generic. Swap-test failures live here.
• Cosmetic. Worth fixing, not costing you a visitor.

Every finding names the page, quotes the line or shows the screenshot, and gives a direction, not finished copy. Then the clearest-moments list.

Running it

From Claude chat, no code. Add Cobalt Capture as a connector once, then ask:

Run a positioning audit of https://example.com

It reads public pages only, so there's no account, no login, and nothing to approve. Cobalt renders each page for the assistant on desktop and on a phone.

From a coding agent. In Claude Code, /mcp__cobalt__positioning_pass https://example.com. Run it from inside your site's repository and the audit compares what the site says against what your docs meant it to say. The studies page has the paste-ready block for Cursor and Codex.

When to run it

• Before a homepage or messaging change ships, on the staging version.
• After a repositioning, to check the new story made it to every page, not just the hero.
• On the sites you compete with, to see what the category's visitors already expect.
• Periodically. A saved review is a baseline; the next run leads with what changed.

Frequently asked questions

What is a positioning audit?

A positioning audit is a structured read of a website's public pages from the point of view of a visitor with no context, deciding in seconds whether the product is for them. It judges what the site communicates, not what the product does: did the page make the visitor understand it, believe it, and know whether it's for them. A page can describe a great product and still fail all three.

What does the audit check?

It starts with a five-second test: from the homepage fold alone, on desktop and then on a phone, what does this do, who is it for, and why believe it. Then it sweeps the key pages, home, product, use cases, pricing, about, docs, and judges each through fixed lenses: who it's for and not for, category and claim and whether the page proves it, the swap test (could a competitor's name replace yours with nothing else changing), jargon the target buyer wouldn't know, consistency between pages, and trust signals near each ask. It also records the clearest moments so nobody regresses them.

Do I need a developer or any setup?

No. Add Cobalt Capture as a connector in Claude chat once, then ask for a positioning audit of your URL in plain words. It reads public pages only, so it needs no account, no login, and no approval step; it runs start to finish on its own. Cobalt takes the screenshots for the assistant, on desktop and phone. From Claude Code or Cursor the same study runs with the agent driving a real browser.

What do I get back?

A review at a shareable link with the verdict at the top: the positioning statement your site actually transmits, in one sentence a cold visitor would repeat to a colleague; then findings ordered by severity (garbled, muddy, cosmetic), each with the page, the quoted line or screenshot, and a suggested direction; then the clearest-moments list. If you run it from your own codebase, it also compares the transmitted positioning against what your positioning docs intend.

How is severity decided?

On outcome, not taste. A finding has to plausibly change whether the target visitor understands, believes, or self-selects. A hero that leads with an acronym the stated audience wouldn't know is a finding. A word the reviewer would have chosen differently is not.

Can I run it on a competitor's site?

Yes. It reads public pages only, never creates an account or submits a form, so any public site is fair game. Running it on the sites you compete with is a fast way to see what the category's visitors are used to reading.