Aiola's crawler directory lists four Alibaba user agents, each with its own robots token. QwenBot, Qwen-User, TongyiBot, and AliyunBot have separate records. Their directory classifications help distinguish automated collection from user-triggered fetching.
What each training record covers
QwenBot, TongyiBot, and AliyunBot are classified as training records with kind crawl in Aiola's directory. Aiola uses that label when collection is automated, rather than a fetch one person triggered. QwenBot's note says the pages support Qwen model and product development. TongyiBot's note ties the identity to Tongyi model development. AliyunBot's note names an Alibaba Cloud identity for that collection.
A robots.txt group naming QwenBot does not name TongyiBot or AliyunBot. Use separate groups when you want different rules for different tokens.
What the user-fetch record covers
Aiola's directory describes Qwen-User as fetching web content in response to a Qwen user request. Aiola classifies it as AI answers because the access happens on demand. The directory kind for this record is fetch. Do not treat a robots.txt block for this token as an authentication boundary.
You can request that all four tokens avoid your site, or name only the training tokens. Keep Qwen-User in its own group when its rule should differ.
Is there an official crawler page?
All four directory records have an empty documentation URL and a last-verified date of July 11, 2026. Alibaba Cloud's AI crawler control guide describes WAF controls for AI traffic. Its examples name other vendors' bots, without naming these four tokens.
That guide does not publish their full user-agent strings or their robots.txt policies. The directory classifications here describe Aiola's records, rather than a vendor-confirmed policy. An empty documentation field does not establish that Alibaba has published no documentation elsewhere.
How do you recognize a request?
Look for the token inside the User-Agent header, using a case-insensitive match as the setup snippets do. Each directory sample is the bare token: QwenBot, Qwen-User, TongyiBot, or AliyunBot. These samples do not supply a full browser-style user-agent string.
All four records set their IP-range type to none. They provide no IP ranges for checking whether a request came from Alibaba. A matching header alone does not authenticate the sender.
When reading a request, separate the token you found from what it proves about the sender. QwenBot names one directory record, while Qwen-User names another. Neither token supplies an IP range you can compare with the source address. Keep that limit in mind when interpreting a matching header.
Read each record separately when deciding which collection you want to allow. A Training classification describes the directory's category for that token. It does not establish a vendor-confirmed robots policy. The same distinction applies to the Answers classification for Qwen-User. Its on-demand label does not turn robots.txt into access authorization.
What belongs in robots.txt?
Give each token its own group, with blank lines for readability. The example requests a whole-site block for each token. It does not establish whether these agents follow robots.txt.
For public access, replace Disallow: / with Allow: /, then add specific Disallow paths as needed. Repeat restrictions in named groups because wildcard rules are a fallback, rather than inherited rules. These matching rules follow the Robots Exclusion Protocol.
User-agent: QwenBot
Disallow: /
User-agent: TongyiBot
Disallow: /
User-agent: AliyunBot
Disallow: /
User-agent: Qwen-User
Disallow: /Publish /robots.txt at the root of each hostname serving the content. Keep existing sitemap entries when adding these groups. Multiple groups matching the same token have their rules combined. Consolidate them carefully rather than assuming the last group wins.
Robots.txt rules are requests to crawlers, not access authorization. Keep confidential pages behind access checks.
Where do the visits show up?
The AI view shows crawl totals, bots, pages, and activity over time. It includes Answers, Indexing, Training, and Other buckets, plus AI-answer referrals, crawler 404s, and discovery-file health. QwenBot, TongyiBot, and AliyunBot land in Training; Qwen-User lands in Answers.
Crawlers do not run JavaScript, so the browser tag cannot see them. One server or edge snippet from the Analytics setup guide posts each matching hit. Only known crawler user agents are stored.
A recorded crawl is not proof that Qwen cited the page. An empty period is not proof that these crawlers cannot reach the site.
Try it in Aiola
Download Aiola free for macOS 12 or later and Windows 10 or later. Hosted Analytics needs Pro or the 7-day trial, and the pricing page lists both. Open Analytics and compare requests from the four Alibaba tokens.
