Sign in Check my site
The Agentic Web Test · how it works

What the test measures, and what it refuses to guess.

People now ask an AI agent before they ever see your site: "who is the best plumber near me, and can I reach them right now?" The agent reads your pages or skips them, then tries to act. The test asks your home page the same three questions a visitor and an agent do, and grades the answers on 100 points. No model writes a word of it.

Free⏱ About a minuteNo sign-inNo AI opinion: every line a measurement
The testHow it worksNext stepsSample report

Three questions, fourteen measured parts

Every part is a fact read off your page as a visitor sees it: the words that arrive before scripts run, the headings, the structured data, a form or a tap-to-call number, hours and an address, the speed on a phone, whether anything moves, whether the photographs are real. Each fact has a rule and a number of points. The letters use school cuts.

Read — can an agent read it?35 points

8Words arrive before scripts run8 when at least 80% of the visible words are in the HTML itself; 5 at 50%; 2 if more than 80 words are; 0 for a shell that only scripts fill.
5Headings outline the page5 for one main heading and two or more section headings; 3 for one main heading alone; 2 for sections with no main heading; 0 otherwise.
8Structured data about the business8 when the page carries schema about the business itself (LocalBusiness, Organization, Service, Product, FAQ…); 4 for site furniture only (WebSite, WebPage); 0 for none.
4A sitemap4 for a sitemap listing five or more pages; 2 for a shorter one; 0 for none.
6An llms.txt6 when /llms.txt answers with a real text file (starts with a heading, not an HTML page returned as 200).
4Open to readers4 unless the page says noindex or robots.txt blocks everything.

Act — can an agent act on it?30 points

10A way to act10 for a form and a tap-to-call number; 6 for one of the two; 0 for neither.
8Hours and an address8 for both in the text; 5 for one; 0 for neither.
6Services named as text6 when four or more sections or menu items name what the business does; 0 when services live only in pictures.
6The trade's must-havesThe industry grade (the things a buyer in that trade looks for: licence, service area, pricing, reviews…) scaled to 6; 3 when the trade is not recognised.

Experience — is it ready for tomorrow's user?35 points

14Fast on a phone14 for Lighthouse 90 or better on a phone profile; then one point per ~4 Lighthouse points down to 0 at 30; 7 when speed could not be measured.
8Something moves or responds8 for a slider, a video or running motion; 5 for three or more controls a visitor can press (tabs, accordions, buttons outside the menu); 3 for one or two; 0 for a wall.
8Real photographs8 for six or more photographs of size; 5 for three to five; 2 for one or two; halved when half or more are stock.
5Built for a phone3 for a viewport meta; 2 for HTTPS.
A
90–100
Agents read it, act on it, and a person stays. Rare.
B
80–89
Ready. The points not earned are the gap between passing and leading.
C
70–79
Partly ready. An agent gets part of the way.
D
60–69
Not ready. Most of the work is still to do.
F
under 60
Agents read past it and people bounce. Every reason is listed, every one is fixable.

What happens in about a minute

1

We open your home page

Exactly as a visitor on a phone would, in our own Chrome. Nothing is submitted, nothing is changed.

2

We read it twice

Once as the HTML arrives, once after scripts run, so we know what an agent gets without a browser and what a person gets with one.

3

We run Lighthouse on a phone

Google’s own speed and search audit, on a phone profile. This is the slow part, thirty to forty seconds.

4

We check the trade’s must-haves

Your industry has a book of what a buyer looks for: licence, service area, pricing, reviews, hours. We read your page against it.

5

We read the brand as it paints

The button colour, the painted colours, the heading and body faces. Measured, not described.

6

We write it up

The grade, the three tests with every signal, the rubric with your points on every line, and everything we found. Cached for the day, so a second look is instant.

Straight answers

Is it really free?

Yes. No sign-in, no card, no trial. The report is yours as a PDF or by email for a first name and an address, and that is the only thing we ask.

Why did my site get an F?

Usually one of three things: the words only appear after scripts run, so an agent reading the HTML sees a shell; there is no form or tap-to-call number; or the page takes more than four seconds to paint on a phone. The rubric names which, line by line.

Does an AI write the report?

No. Every line is a measurement with a rule. The one place a model is used anywhere in Think Bright Sites is when it builds your new site, and a person reviews that before it goes live.

Is my report public?

Only if you publish it. The report page and the badge exist when you tick the box on the report form; until then the result is yours alone. Checks are cached per site for a day.

Can I run it again?

Any time. A check is cached for the day, so a second run on the same day shows the same result instantly; tomorrow it reads the site fresh.

What does it not check?

Inner pages, your Google Business Profile, reviews on other sites, your ads. It grades the home page an agent and a person land on first, which is where most sites lose the customer.