Agentic IntentAgentic Intent · web infrastructure
What can an AI assistant actually read on your site?
Usually less than you think, and some sites it cannot reach at all. Thirty seconds, free, no sign-up, and the result is yours.
Free · no email to see the result · we honour robots.txt
The problem
Almost nobody publishes anything built for an assistant to read.
We scanned all 500 of the largest US public companies. 297 publish nothing written to be quoted, 35 serve a file that only looks like an answer, and 64 block AI crawlers by name, the one case where an assistant genuinely cannot reach them. The rest of the time it reads the page like anything else and fills the gaps from a reseller, a review site or a competitor. The index and the method →
Scanned recently
Anyone can run one. Results are public.
What we build
Five layers. One engagement.
They do not work apart. A fast site nobody can find is a brochure; a readable site that takes nine seconds has already lost the person.
Site Rebuild
The site itself, done again.
Agent Readiness
What you publish for an assistant to read.
Onsite Assistants
Your own agent, on your own data.
Channel Management
An assistant that recommends you is a channel.
SEO
The channel that still sends most of the traffic.
AEO
Answer-engine optimisation.
Performance
Measured where customers are.
What it is for
Three jobs a website can have.
Most companies are one of these clearly and a second one quietly. The engineering is shared. The outcome is not.
Agentic Commerce
The assistant brings a cart.
- Retail & e-commerce
- Furniture & home
- Grocery
- Direct-to-consumer
AI Intake
The assistant brings a screened lead.
- Law firms
- Medical & dental
- Car dealerships
- Home services
AI Visibility
The assistant quotes you, not a competitor.
- Consumer brands
- Pharma
- Health systems
- Universities
Proof
Rebuilt. Measured. Running.
One engagement: a twenty-store retailer’s entire public site, 14,413 pages of it, rebuilt as a working demo with an assistant on it that knows every priced product and hands a real basket to their checkout.
100, 100 and 100 on Google’s own test.
Home, product and category pages, on a computer, on every run. On a phone the same pages score 99, 100 and 100.
Those pages paint in 367 ms, 523 ms, 523 ms on home, product and category. The site they replaced took 25.7 seconds to paint its home page on a phone.
| Home page, on a phone | Before | After |
|---|---|---|
| Performance score | 31 | 99 |
| Largest paint | 25.7 s | 1.6 s |
| Page weight | 20 MB | 0.5 MB |
| Requests | 359 | 25 |
Google Lighthouse, 28 September 2026. Phone figures are the median of three runs over the public internet; the paint times above are the same on a computer. Client names on request.
How we work
Ninety days, three phases. Then ongoing care.
The rebuild
The site itself: every page, every product, connected to the systems you already run. Measured against the site it replaces, on the public internet.
The agent-ready site
Written so outside models can read, price and cite your catalog: the file at the door, the machine-readable twin of every page, the facts they quote back.
The assistant
Your own agent on the site, grounded in your catalog or your intake criteria, replacing whatever chat box is there now. It fills a real cart or files a real record.
Care and measurement
Kept current, kept fast, kept readable. One report every month of what each channel produced, with the model cost shown at cost.
Start with what your site is meant to produce.
Tell us whether it sells, qualifies or explains, and we will tell you what is standing in the way. The first look costs nothing.