You built the homepage yourself. You know what the product does, so every line reads clearly to you. The problem is you are the wrong reader. The person you need to convince landed from a cold link, gave you eight seconds, and left with a guess about what you sell. A positioning audit tells you what that guess was.
Here is the whole thing worked through on a made-up but realistic site, so you can see the inputs and the actual output instead of a description of them.
The input: one URL, said in plain words
The founder runs a tool called Threadline that turns Slack threads into a decision log. The homepage headline reads "Threadline keeps your team aligned." Below it: a hero screenshot, three feature cards, a pricing link, and a "Book a demo" button.
To run the audit, the founder added Cobalt to the AI they already use (Settings, Connectors, Add custom connector, paste the MCP URL) and typed one line: Run a positioning audit of threadline.com. No code, no prompt writing. Cobalt renders the public pages itself, so there was nothing to install and nothing to babysit. The study runs start to finish and comes back as a shareable review with screenshots of each page, findings ranked by cost, and a verdict at the top. If you want the full scope of what it checks, the positioning audit study lays out desktop and phone, homepage and key pages.
The output: findings ranked by what they cost
The verdict at the top was blunt: a cold visitor cannot tell who Threadline is for or what it replaces. Three findings carried that verdict.
1. The headline describes a feeling, not a job
"Keeps your team aligned" could sit on a project tracker, a chat app, or a calendar. The audit flagged it against the first question a cold visitor asks: what is this, in one read. The screenshot of the hero showed the headline with a pin on it and the note that nothing in the first viewport names Slack, decisions, or logs. The visitor's guess after eight seconds: "some kind of team chat thing." Wrong, and expensive, because that guess sends them away.
2. The audience is never named
The three feature cards said "Smart summaries," "Instant search," and "One source of truth." The audit noted that none of them says who this is for. A five-person startup and a 200-person ops team read those cards completely differently. Naming the reader ("for teams that make decisions in Slack and lose them by Friday") does more work than any feature card. The finding came with the exact card text quoted, so the copy fix is obvious.
3. The claim has no proof next to it
"One source of truth" is a claim. The audit checked whether the page proves it and found nothing: no before-and-after, no example log, no screenshot of an actual decision captured. The hero image was a blurred generic dashboard. A cold visitor has no reason to believe the claim, and the page gives them none. This ranked lower than the headline, but it is the thing that loses the visitor who almost believed you.
The phone view changed one finding's rank
On desktop the "Book a demo" button sat beside a "Start free" link. On the phone render, the audit caught that "Start free" dropped below the fold while "Book a demo" stayed pinned at the top. For a solo founder whose buyers want to try before they talk, that reorder quietly pushes the cheaper, faster path out of sight. You would not catch this reading your own site on a laptop. The audit caught it because it checked both, and the screenshot proved it. If pricing clarity is your worry specifically, the pricing page teardown is the sibling study to run next.
What you do with the review
The review arrives as a link anyone can open. The founder here handed it to a copywriter, who rewrote the headline and two cards straight from the quoted findings. If you work with a writer, the format is built for exactly that handoff, as covered in positioning audit notes a copywriter can rewrite from. The same review is also available as clean markdown, so if you ship with an AI coding agent, it can read the findings and the screenshots directly.
A quick word on cost, because honesty is the point. Running the study and reading the review is free to try, with no password or card. The review stays editable for seven days; after that it goes read-only for you, though the shared link keeps working for whoever you sent it to. Keeping every review editable past that window is the $5 a month Pro plan, cancellable any time, and the full pricing spells out the yearly option.
The point of running one is not to be told your site is bad. It is to read your homepage the way a stranger does, with the stranger's questions attached to the exact pixels that failed to answer them. Pick your own URL and run the audit before you rewrite a single line.