TL;DR
Write down what you sell, to whom and why it matters, then ask an AI assistant to read your site and answer the same questions. Where its answers differ from yours, your page is unclear, not the reader. With fewer people clicking through from search, the visitors you do get need to understand faster.
Write down, in one sentence each, what you sell, who it is for and why it matters. Then ask an AI assistant to read your site and answer the same three questions without your help: wherever its answers differ from yours, your page is unclear. If a model reading every word cannot work out your project, a visitor skimming for a few seconds will not either.
Why clarity matters more than it did
Your site is your first salesperson, and it works alone. If it does not say what you sell, to whom and why, nobody will guess.
It also gets fewer chances than before. Pew Research found that Google users who saw an AI summary clicked a traditional search result in 8% of visits, against 15% for those who did not (Pew Research, 2025). Fewer people arrive, and the ones who do have to understand quickly.
A worked example: what fewer clicks cost you
Take the Pew rates as a rough illustration, not a forecast for your site. Say 2,000 searches a month could lead to your page, and assume every click lands on you.
- Without an AI summary: 2,000 × 15% = 300 visits
- With an AI summary: 2,000 × 8% = 160 visits
Now assume 3% of visitors sign up today.
- 300 × 3% = 9 signups a month
- 160 × 3% = 4.8, so about 5 signups a month
To get your 9 signups back from 160 visits, the page has to convert 9 ÷ 160 = 5.6% of visitors. That is almost double the conversion rate, with nothing else changed. You will not get there with a nicer button. You get there when more of the people who land on the page understand it on the first read.
These numbers are assumptions chosen for the example. The point is the shape of the math: when traffic shrinks, every unclear sentence costs more.
The test, step by step
1. Write your own answers first. Before you open any tool, write down four things, one sentence each:
- What a customer buys from you
- Who it is for
- The bet behind the project: why this, why now
- Your headline, the line you want people to remember
Doing this first matters. If you read the model's answer before writing yours, you will agree with whatever sounds close enough.
2. Give the model your pages, nothing else. Paste the text of your homepage, pricing page and one feature page into an AI assistant, or let it read the URLs if it can. Do not add any context about your project.
3. Ask the same four questions. "From these pages only, what does a customer buy? Who is it for? What is the bet behind this project, in one sentence? Quote, word for word, the sentence that best sums it up."
4. Score each answer. Compare with what you wrote:
- Match: same meaning, even in different words
- Close: right area, wrong detail (wrong audience, wrong outcome)
- Miss: wrong, or the model says it cannot tell
Four matches is a clear page. One miss is worth fixing before you spend anything on traffic. Two or more, and the page is describing a different project from the one in your head.
5. Fix the page, not the prompt. Rewrite the sentences that caused the misses, then run the test again in a fresh conversation, so the model does not remember the previous answer.
Reading the results
The quote question is often the most useful one. If the model picks a sentence buried in your FAQ instead of your headline, your headline is not doing its job. If it cannot find any single sentence that sums you up, you do not have one yet.
A "close" on the audience usually means your page lists several types of user without saying who comes first. A miss on the bet usually means the page describes features and never says what changes for the customer.
Your pricing page counts too. If the model cannot say what a customer buys, check what the plans actually name. See how to price a micro SaaS with zero customers if that page is still a blank.
What this test does not cover
It measures whether your pages can be understood, not whether the offer is good. A perfectly clear page for something nobody wants still converts nobody. Whether the offer holds up is a separate set of questions, laid out in what to do after building an app with AI.
It also judges words only. Design, load speed, screenshots and trust signals all affect real visitors, and a text reading misses them.
And it is weak at spotting what is absent. A model reading your pages will describe what is there. It will rarely tell you reliably that a whole section is missing, because it cannot know what you meant to say. Treat its answers as a mirror of what your site says, not as an audit of what it should say.
How Banast does it
The free Banast pre-audit starts with exactly this step, under the heading "What Banast understood". It reads up to 20 of your pages, sitemap first, sorts them, then extracts the facts with the page each one came from. It gives you your pitch, your hook quoted word for word, your bet in one sentence, and three questions on strategy, messaging and marketing.
You answer "That's right" or you correct it, and the rest of the report starts from your version. The questions are always built from what your site says, never from what it seems to leave out, for the reason above: a crawl sees absence badly.
FAQ
Why use an AI instead of asking friends?
Friends know you and fill the gaps for you. A model only has your pages, like a stranger arriving from search. Use both if you can, but the model is available tonight and does not try to be nice.
The AI got it wrong. Should I fix the prompt?
No. Fix the page. If you have to explain your product in the prompt for the model to get it, a visitor would need the same explanation, and nobody is there to give it.
Does this replace testing with real users?
No. It catches pages that cannot be understood at all. Whether the people who understand it also want it is a separate question that only real buyers answer.
Adrien Lebas
Founder of Banast, building it alone from Avignon. Writes about what it takes to make a living from what you build.