01Notes · 20 August 2026
Technical SEO for agents: making a site readable, not just crawlable
Quick answer
A crawler reads your page and stores it. An agent is trying to finish somethingfind a price, compare two vendors, reach a form. It needs your content to be usable, not just indexed.
The good news: almost everything that helps an agent helps a person. If you find yourself building a second site for machines, something has gone wrong.
Key takeaways
- Readable is not the same as usable. A page can be indexed perfectly and still be useless to something trying to act on it.
- JavaScript-only content is the most common blocker, and the most common finding in our audits.
- State facts as text, not as pictures. A price in an image is invisible to everything except a person looking at it.
- Real HTML beats visual arrangement. Headings that are headings, lists that are lists, buttons that are buttons.
- Test by turning JavaScript off and completing your own key task. Ten minutes, and it is conclusive.
- Do not build for agentic commerce yet. Track it. The standards are still moving.
An agent needs more than a crawler does
A crawler has a simple job. Fetch the page, store the content, move on. Success is being readable.
An agent has an errand. Someone asked it to find the cheapest option, or check whether a tool integrates with something, or start a trial. Success is completing the task, and that is a much higher bar than being readable.
A page can be indexed flawlessly and still fail an agent completely: the price is in an image, the integration list loads after a click, the specification is in a PDF behind a form. Every one of those is fine for a crawler and fatal for an errand.
| On your page | A crawler | An agent trying to act |
|---|---|---|
| Price shown in an image | Indexes the page fine | Cannot read the price. Task fails |
| Specs load after a click | May index the initial state | Never sees them |
| Content renders only via JavaScript | Google may render; others may not | Often blank |
| Key facts only in a PDF | Indexes the PDF separately | Usually gives up |
| Divs styled to look like buttons | Does not care | Cannot tell what is clickable |
| Plain HTML, stated facts, real forms | Fine | Fine. This is the whole answer |
The last row is not a technique. It is what a well-built site already looks like, which is why this rarely needs a separate budget line.
The five things that break it
- Critical content behind JavaScript. The most common finding in our audits, and usually nobody knew.
- Facts as images. Prices, specifications, comparison tables rendered as graphics.
- Content behind interaction. Tabs, accordions and modals that load on click rather than being present and hidden.
- Non-semantic markup. Divs pretending to be headings, buttons and lists. A machine reads structure, not appearance.
- Facts stranded in PDFs that have no HTML equivalent.
None of these is exotic and none needs new technology to fix. All five sit inside ordinary technical work.
A ten-minute test
- Turn JavaScript off in your browser settings.
- Try to complete your own key task. Find the price. Reach the contact form. Compare two plans.
- Note where you get stuck. Every point where you cannot continue is a point where an agent stops too.
- Turn JavaScript back on and view source on your most important page. Search for your price as text. If it is not there, it is not readable.
Ten minutes, no tools, no spend. It produces a shorter and more honest list than most paid audits of this kind.
Leave this for later
There is a growing vocabulary around agentic commerce — protocols for agents to transact, structured product feeds built for machine purchase, agent-specific checkout.
We track it and we do not sell it. The standards are still moving, and building against a moving standard produces something you maintain for years and nobody uses. When a standard settles and your buyers are actually using it, that is the moment.
Until then, the work that helps is the work in the list above, which also helps every human visitor you already have.
What this means for you
- Do the JavaScript-off test this week. Ten minutes, and it is the most honest audit available.
- Get your prices and specs into text. If it matters, it should not be a picture.
- Do not commission agentic-commerce work yet. Track the standards instead.
02Common questions
Questions this raises
What is the difference between a crawler and an agent?
A crawler reads a page and stores it. An agent is trying to complete something — find a price, compare two options, book a slot. It needs your content to be usable, not just readable.
Do we need a separate agent version of our site?
No. Almost everything that helps an agent helps a person: clear structure, real HTML, stated facts, working forms. If you are building a second site you have taken a wrong turn.
Does JavaScript break agent access?
It can. Content that renders only after JavaScript may never be seen by systems that do not execute it. This is also the single most common technical finding in our audits.
What about llms.txt?
Google names it as unnecessary for its generative features. It is not an agent-readability measure; it is a file that looks like one.
How do we test agent readability?
Disable JavaScript and try to complete your own key task. If you cannot find the price or reach the form, neither can a system that does not render.
Is this the same as accessibility?
Overlapping and not identical. A site built properly for screen readers is usually far ahead on agent readability, because both depend on real semantic HTML rather than visual arrangement.
Should we build for agentic commerce now?
Track it, do not build for it yet. Standards are still moving, and building against a moving standard is how you end up maintaining something nobody uses.
Sources
Every figure on this page comes from one of the following. Where a claim is ours rather than a source’s, it carries an evidence label instead.
- Google Search Central — Optimizing your website for generative AI features (15 May 2026, updated 10 July 2026), structured data documentation, and the 2026 deprecation of FAQ rich results. Primary
- Rankability — State of AI Search, 1,645 citation observations across 31 topics, 2026. Source for the finding that no schema type showed a consistent independent citation advantage, and for the first-100-words and query-breadth findings. Independent study
- Searlixa sampling across client engagements, labelled Observed with sample and date. Our own
Ask an assistant
Hand this page to an assistant and ask it to summarise or challenge the argument. If it gets something wrong about us, that is itself a useful finding.
→Next step