Scanning live. A first-ever scan of a domain runs against the site and takes about ten seconds.
0.0s elapsedAgentSpeed tests how easily automated agents can discover, understand, and act on irs.gov. The score weights five categories of machine-readability; the ticks on the arc are the 37 individual checks behind it.
Send this to whoever edits the site. The link shows the score and grade as a preview card, and it stays current: it re-reads the latest scan rather than freezing today’s number.
No JSON-LD blocks found. Add a <script type="application/ld+json"> with a schema.org type for this page.
Show me exactly what to doJSON-LD is a small block of structured facts (name, what you sell, prices) embedded invisibly in your page’s HTML, written in a vocabulary (schema.org) that machines share. Think of it as the machine-readable caption for the page humans see.
An agent reading prose has to guess which number is the price and which name is the brand. JSON-LD removes the guessing: typed facts it can quote directly. Pages without it get summarised from inference, which is where wrong prices and wrong names come from.
Get it in your inbox now, plus a heads-up whenever irs.gov’s agent-readiness score changes. Free, unsubscribe anytime.
Prefer we just watch it for you? Monitor rescans weekly and emails what changed, $29/mo. Start monitoring →
Generated, ready-to-ship files for the gaps above. Copy them, or download and drop them into your repo.
Weakest on this scan: structured data, 0/100 — facts like price and availability are missing or unreliable in machine-readable form, so agents are left to guess.
This graded your homepage. Your pricing, docs, and checkout live on other pages: the ones assistants actually quote when they recommend you. A 63/100 here means agents are already missing things on the page you polish most. The $29 Deep Scan grades your homepage plus up to 10 more pages we discover, against all 37 checks, then emails a PDF with every failing check and its exact fix, ranked by what costs you the most visibility. One-time, delivered in minutes, no subscription.
Embed this badge on your site. It links back to this live scan and updates on every rescan.
<a href="https://agentspeed.com/scan/irs.gov"><img src="https://agentspeed.com/badge/irs.gov" alt="AgentSpeed agent-readiness score" height="56"></a>Get this report by email, then a heads-up when irs.gov’s agent-readiness actually changes: a check regressing, or the composite dropping.
Two things never trigger an email: a score change caused by us publishing a new rubric, because that is a different instrument rather than a change to your site, and movement explained only by timing-derived checks, which shift run to run on a site nobody touched.
Want it on your own schedule, with Slack or a webhook instead of email? Score-drift monitoring is on every paid plan.
This score says whether an AI agent could read irs.gov. AI traffic analytics says whether one actually came, and which pages turned it away.
Running AI agents of your own? AgentSpeed also monitors them in production: every run, its latency, cost, and failures, with alerts when something breaks. Free for 10,000 events a month, no credit card.
This score is a measurement of machine-readability for automated agents under the published rubric 2026.11.0. AgentSpeed does not assess security, legitimacy, privacy, or financial trust.
Paste into Claude Code, Cursor, or any assistant with your repository open. It carries this finding, the steps, and the shortcuts to avoid.
No HTTP Link: response headers. Emit canonical/alternate/describedby relations so HEAD-only or stream-rendering agents get them without parsing HTML.
Show me exactly what to doHTTP responses can carry a Link header: the same canonical and alternate-language information your HTML declares, but delivered in the response envelope itself. Agents that only send a HEAD request, or that decide what to do while the page is still streaming, read the header without parsing any HTML. Yours is missing (or carries none of the relations agents use: canonical, alternate, describedby).
An agent triaging fifty URLs doesn’t want to download and parse fifty pages to learn which are duplicates of which. The Link header answers at the cheapest possible layer. Sites that provide it get correctly de-duplicated and correctly language-routed even by the most minimal fetchers.
Paste into Claude Code, Cursor, or any assistant with your repository open. It carries this finding, the steps, and the shortcuts to avoid.
<title> is 78 characters; keep it under 70 so it isn't truncated.
Show me exactly what to doThe <title> tag is the page’s name: the text in the browser tab. Yours is missing, empty, or generic (e.g. just "Home"), so machines have no usable name for this page.
The title is the single most-quoted identifier for a page; agents cite it verbatim when referencing you. A missing one gets replaced with a guess assembled from headings, which amounts to your page’s name being written by someone else.
Paste into Claude Code, Cursor, or any assistant with your repository open. It carries this finding, the steps, and the shortcuts to avoid.
Page looks like Article|BlogPosting|NewsArticle; missing JSON-LD type(s): Article|BlogPosting|NewsArticle.
Show me exactly what to doYour structured data exists but only covers the basics (say, Organization) while the page is really a product page, an article, or an FAQ. The machine-readable caption is missing the part that describes what this page actually is.
An agent looking for "running shoes under $100" matches against Product markup with offers. A product page carrying only Organization markup is invisible to that query shape: the agent knows who you are but not what this page sells.
Paste into Claude Code, Cursor, or any assistant with your repository open. It carries this finding, the steps, and the shortcuts to avoid.
Visible text is 8.7% of HTML weight (11743/135423 bytes). 3% of the document is inline script, 88% is markup.
Show me exactly what to doOf everything your server sends for this page, very little is actual readable text. The rest is code, styling, and markup wrapper. Agents fetched a lot of bytes and found few words.
Agents work with retrieval budgets. A page that is 2% text either gets skimmed (and mis-summarised) or skipped. More signal per byte means more of your actual message survives into the agent’s summary.
Paste into Claude Code, Cursor, or any assistant with your repository open. It carries this finding, the steps, and the shortcuts to avoid.
Heading issues: no <h1>.
Show me exactly what to doHeadings (H1, H2, H3…) are the page’s outline. Yours skip levels, repeat H1s, or use styled text instead of real heading tags, so the outline machines extract is scrambled.
Agents use the heading tree to navigate long pages and to decide which section answers a question. A broken tree means quotes get attributed to the wrong section, or sections are missed entirely.
Paste into Claude Code, Cursor, or any assistant with your repository open. It carries this finding, the steps, and the shortcuts to avoid.
One prompt covering the fixes above that have a step-by-step plan, in the same priority order, with the same guardrails.
This scan shows whether an agent can read the site. One free probe asks a real assistant a real buying question and records who it cited.