SmartHomeWorkflowBot

If you found this page from a line in your server logs, you're in the right place.

SmartHomeWorkflowBot re-checks the pages our own published facts cite. When a product page on this site says a device works with Alexa, or costs what it costs, or needs a subscription, that claim carries a link to the manufacturer page it came from. The bot revisits those specific pages to notice when they change — so we can correct ourselves before a reader acts on something stale. That's the entire job.

What it does not do

• It does not copy your content. The bot compares a page against what we recorded last time and flags a difference for a human to read. We publish our own words about products, not yours.
• It does not crawl your site. It has a list of specific URLs — the ones our records already cite — and it fetches those. It does not follow links looking for more.
• It does not train anything on your pages, and here is the honest detail rather than a comfortable one: when a page has changed, we do send that page's text to a language model to describe what changed against what we recorded, and a person reads the answer before any of our pages move. Unchanged pages never get that far. Nothing is used to train a model, and nothing is republished.

How to block it

The user-agent token is SmartHomeWorkflowBot. Put this in your robots.txt and we will stop:

User-agent: SmartHomeWorkflowBot
Disallow: /

We check robots.txt before every page request, and a disallowed path is skipped and logged — never worked around. Three details worth stating because they cut in your favour and you shouldn't have to take them on faith:

• If your robots.txt is unreachable, we treat that as a no and skip the page. An unanswered question is not a yes.
• If you publish a Crawl-delay longer than ours, yours wins.
• If you block AI crawlers, that includes us — even though our name isn't on your list. Plenty of sites now disallow GPTBot, ClaudeBot, Google-Extended and their neighbours while leaving * open. Read strictly, that permits us: we're none of those. We treat it as a no anyway. You've said what you want from agents like ours, and needing you to also guess our bot's name first would be a technicality — not something a site that sells itself on honest sourcing gets to lean on.

You can also just return a 403. A 403 is a policy statement, and we record it and stop asking. We do not rotate user-agents, use proxies, or spoof headers to get around a refusal — not as a courtesy, but as a rule the code enforces.

Sometimes it uses a browser

A lot of product pages now send an empty shell to a plain request and build the real page with JavaScript. When that happens we load the page in a real browser instead, read the text a visitor would have seen, and stop there — one page, one load, no scrolling, no forms, nothing dismissed. It is the same single polite request wearing the same user-agent, just rendered. Everything above still applies to it, and that is the part worth saying plainly: we check your robots.txt first and skip a disallowed path, and if you have refused us with a 403 or a bot-check page, a browser is exactly the tool that could get around that, so the code refuses to run it at all against a host that has said no.

One more thing we do, described plainly because it sits close to a line: when a page puts detail behind a tab or an accordion, we may open it, the same way a visitor reading that page would. That text was already sent to us in the page itself, opening it asks your server for nothing further, and it goes nowhere a click wouldn't. What we will not do is dismiss anything to get at content — a cookie banner, a consent dialog, a paywall or a sign-in prompt is a decision your site has made, and clicking past one is not reading a page, it is working around it. We do not fill in forms, sign in, or leave any trace a visitor's own click wouldn't. And none of this touches the rules above: if you have refused us, or your robots.txt says no, a browser is exactly the tool that could get around that, so we don't point one at you at all.

How often, honestly

Less than you'd expect, because being cheap and being polite happen to be the same thing here:

• One request at a time. The whole run is a single sequential pass — the bot never opens parallel connections, to your site or to anyone's.
• A courtesy pause between requests to the same host, two seconds by default, and longer if your Crawl-delay says so.
• Conditional requests first. We send the validators from last time, so if nothing changed you answer 304 Not Modified and send no body at all. That is the outcome we're hoping for.
• One robots.txt fetch per host per run, cached in memory for that run.
• Price-only sources are checked monthly, not daily — scheduled runs skip them outside the first week of the month.
• If you rate-limit us with a 429, we honour Retry-After once. If waiting has already proven not to help with your host, we stop for the day instead of pushing.

Why this exists

The promise this site makes is that its facts are true, not merely numerous — and a fact with a date on it decays unless something goes back and looks. That's what the bot is: the going-back-and-looking, done on a schedule instead of whenever someone remembers. The longer version is on how we work.

Get in touch

If the bot is causing you trouble, or you'd rather we cite a different page, or you want it gone — tell us and we'll fix it. Manufacturers: if we've recorded something about your product incorrectly, our corrections policy is the fastest route, and we treat your own documentation as the strongest evidence there is.