Check a landing page with Claude: what a prompt does and what your audience adds

How to check a landing page with Claude alone

• Copy into the chat: paste the text of the page. The URL is enough only if your Claude can fetch web pages.
• Set the frame: who reads the page, where do visitors come from (search, ad, referral), and what should the next step be?
• Ask for obstacles: where does a reader stop? What does someone without prior knowledge not understand? Which objection stays open?
• Ask for evidence: for every point, the passage on the page, quoted.
• Second check: ask the same question in a new chat and compare. What shows up only once is a hint, not a finding.

A sample prompt to adapt: “Here is the copy of our landing page. It is aimed at [audience]. Visitors come via [search term or ad]. The next step is [demo, trial, enquiry]. Name the three passages where a reader from this audience is most likely to stop, each with a quote and the reason. Do not give scores.”

Where a prompt reaches its limits

• No audience with evidence: Claude judges from general knowledge about buyers. What really holds your buyers back is in reviews, forums and conversations, not in the prompt.
• Agreement: Language models tend to agree with their users. A study by Anthropic calls this “sycophancy” and finds it in several widely used assistants. So ask for problems, not for approval, and check whether the criticism also comes when you do not ask for it.
• Estimating instead of measuring: contrast, tap targets on mobile and page length can be measured. For each such statement, check whether it was measured or estimated from text or an image.
• No comparison: rating one page on its own does not tell you whether a new version lands better. That takes two versions and the same readers.

What 99takes in Claude adds

First, your audience: 10 to 14 personas with a wide spread, backed by public customer voices with their source where they exist. You set it up once for 15 credits, and then it is there for every check. How that works is under audience research.

Analysing a page has several parts. Structure and reading flow are measured (1 credit). The impact on desktop and mobile costs up to 12 credits, plus 2 to 3 for redesign suggestions. Then come the copy from the personas’ point of view, the comparison with competitors and a plan of what to change first. More under what 99takes checks.

If you want a new version to compete with today’s, 99takes writes it. You see it before the test, then the personas read both blind (buyer simulation).

The honest limit: this is a simulation before launch, not a market measurement. How often the jury was right is shown by the hit rate, published from ten decidable results.

What it looks like in the chat

You write in Claude: “Analyse https://[your-page] with our audience.” Claude calls analyze_page for this. 99takes first names the price per part, at no charge, and only starts after your yes. The result comes in the same conversation: what the page costs today, where readers drop off and what to change first.

If you want to test a new version right away: “Run an A/B test for https://[your-page].” Then test_page guides you through scope, price and the written version before the personas read. More questions to copy under example questions.

Connect 99takes to Claude

In claude.ai you add 99takes as a connector, without a key. In Claude Code you run one command: claude mcp add --transport http 99takes https://srcblvgafgkdeaaowhfu.supabase.co/functions/v1/mcp. Then check the connection with “Are you connected to 99takes? How many credits do I have?”. All steps under setup.

Frequently asked questions