Home › Guides › AI Fetcher IP Ranges
✓ checked 17 October 2026 · sources: OpenAI, Google and Perplexity IP-range files, Internet Archive Wayback Machine, iptoasn.com · method · caveats · download the data
The IP list OpenAI publishes for ChatGPT-User, the agent ChatGPT sends to a web page when a user asks it to, now covers 36,496 IPv4 addresses, 13.5 times the 2,704 in its GPTBot crawler list. Google’s list for Google-Agent, its newer agent for browsing and acting on a user’s behalf, covers 70,016 addresses, 13.8 times its Googlebot list. In both companies’ files, address space for fetches started by a user now outweighs address space for crawling.
ChatGPT-User addresses 36,496Google-Agent addresses 70,016ChatGPT-User, a year earlier 2,453
Headline findings
- ChatGPT-User’s list grew 14.9-fold in a year, mostly from one block. The Wayback Machine’s copy from 13 October 2025 lists 167 prefixes and 2,453 addresses; today’s file, dated 7 October 2026, lists 234 prefixes. The jump came in June 2026, when the total went from 3,952 to 36,816 with the addition of 9.129.0.0/17, a single block of 32,768 addresses that is 89.8% of today’s total. The other 233 entries are /28 blocks of 16 addresses. The earliest capture, from November 2024, had 15 prefixes and 240 addresses.
- GPTBot’s list has shrunk. It peaked at 3,472 addresses in lists dated from 30 October 2025 and now has 18 prefixes and 2,704 addresses. OAI-SearchBot, which OpenAI uses for search, has 2,560. iptoasn.com maps every address in all three OpenAI lists to Microsoft’s AS8075.
- Google-Agent’s list is one /16 and a few small blocks. Of its 12 IPv4 prefixes, 136.122.0.0/16 alone holds 65,536 addresses. iptoasn.com maps 67,840 addresses (96.9%) to AS396178, named G-CLOUD-AGENT-TRAFFIC; 2,048 were not routed in its data. Google says Google-Agent “is used by agents hosted on Google infrastructure to navigate the web and perform actions upon user request”. The file’s first Wayback capture, in March 2026, already held 67,840 addresses; it has had the same prefixes since May.
- Both companies say robots.txt may not apply to these fetches. OpenAI says that because ChatGPT-User actions are initiated by a user, “robots.txt rules may not apply”; Google says its user-triggered fetchers “generally ignore robots.txt rules”. Google’s two older user-triggered lists hold 17,088 addresses (fetches by Google Cloud customers’ apps, up from 10,112 in April 2023) and 8,064 (Google-run fetchers). Perplexity’s two lists are far smaller: 18 addresses for PerplexityBot and 14 for Perplexity-User.
Published address space, by list
| Operator | List | Type | List dated | IPv4 prefixes | IPv6 prefixes | IPv4 addresses | Main networks (iptoasn.com) |
|---|---|---|---|---|---|---|---|
| User-triggered agents (Google-Agent) | user-triggered | 16 Oct 2026 | 12 | 8 | 70,016 | AS396178 G-CLOUD-AGENT-TRAFFIC; AS0 Not routed | |
| OpenAI | ChatGPT-User | user-triggered | 7 Oct 2026 | 234 | 0 | 36,496 | AS8075 MICROSOFT-CORP-MSN-AS-BLOCK |
| User-triggered fetchers (users) | user-triggered | 16 Oct 2026 | 534 | 534 | 17,088 | AS396982 GOOGLE-CLOUD-PLATFORM; AS139190 GOOGLE-AS-AP Google Asia Pacific Pte. Ltd. | |
| User-triggered fetchers (Google) | user-triggered | 16 Oct 2026 | 252 | 252 | 8,064 | AS15169 GOOGLE | |
| Common crawlers (Googlebot) | crawler | 16 Oct 2026 | 170 | 147 | 5,088 | AS15169 GOOGLE; AS396982 GOOGLE-CLOUD-PLATFORM | |
| Special crawlers | crawler | 16 Oct 2026 | 136 | 136 | 4,352 | AS15169 GOOGLE | |
| OpenAI | GPTBot | crawler | 22 Sep 2026 | 18 | 0 | 2,704 | AS8075 MICROSOFT-CORP-MSN-AS-BLOCK |
| OpenAI | OAI-SearchBot | crawler | 2 Jan 2026 | 39 | 0 | 2,560 | AS8075 MICROSOFT-CORP-MSN-AS-BLOCK |
| Perplexity | PerplexityBot | crawler | 7 Feb 2025 | 8 | 0 | 18 | AS14618 AMAZON-AES |
| Perplexity | Perplexity-User | user-triggered | 17 Oct 2025 | 4 | 0 | 14 | AS14618 AMAZON-AES |
ChatGPT-User’s list over time
| Wayback capture | List dated | IPv4 prefixes | IPv4 addresses |
|---|---|---|---|
| 2024-11-05 | 9 Oct 2024 | 15 | 240 |
| 2025-01-13 | 6 Jan 2025 | 24 | 384 |
| 2025-04-16 | 15 Apr 2025 | 37 | 592 |
| 2025-05-01 | 30 Apr 2025 | 61 | 976 |
| 2025-06-03 | 16 May 2025 | 84 | 1,344 |
| 2025-07-02 | 6 Jun 2025 | 103 | 1,648 |
| 2025-09-10 | 27 Aug 2025 | 139 | 2,224 |
| 2025-10-13 | 10 Oct 2025 | 167 | 2,453 |
| 2025-12-10 | 25 Nov 2025 | 193 | 3,088 |
| 2026-01-13 | 7 Jan 2026 | 208 | 3,328 |
| 2026-02-11 | 9 Feb 2026 | 224 | 3,584 |
| 2026-03-12 | 10 Mar 2026 | 234 | 3,744 |
| 2026-04-04 | 25 Mar 2026 | 238 | 3,808 |
| 2026-05-24 | 21 May 2026 | 247 | 3,952 |
| 2026-06-20 | 16 Jun 2026 | 254 | 36,816 |
| 2026-07-03 | 24 Jun 2026 | 254 | 36,816 |
| 2026-08-11 | 10 Aug 2026 | 256 | 36,848 |
| 2026-09-05 | 4 Sep 2026 | 207 | 36,064 |
| 2026-10-02 | 25 Sep 2026 | 230 | 36,432 |
GPTBot’s list over time
| Wayback capture | List dated | IPv4 prefixes | IPv4 addresses |
|---|---|---|---|
| 2023-10-22 | 15 Sep 2023 | 11 | 656 |
| 2023-12-10 | 30 Nov 2023 | 2 | 512 |
| 2024-03-18 | 30 Nov 2023 | 2 | 512 |
| 2024-06-12 | 30 Nov 2023 | 2 | 512 |
| 2024-08-12 | 30 Nov 2023 | 2 | 512 |
| 2024-11-06 | 30 Nov 2023 | 3 | 768 |
| 2024-12-30 | 30 Nov 2023 | 6 | 1,168 |
| 2025-01-01 | 30 Nov 2023 | 6 | 1,168 |
| 2025-02-11 | 30 Nov 2023 | 7 | 1,184 |
| 2025-03-15 | 30 Nov 2023 | 7 | 1,184 |
| 2025-04-13 | 11 Mar 2025 | 6 | 1,168 |
| 2025-06-05 | 22 May 2025 | 8 | 1,680 |
| 2025-07-02 | 22 May 2025 | 8 | 1,680 |
| 2025-08-03 | 22 May 2025 | 8 | 1,680 |
| 2025-09-10 | 22 May 2025 | 8 | 1,680 |
| 2025-10-25 | 3 Oct 2025 | 15 | 2,704 |
| 2025-11-18 | 30 Oct 2025 | 21 | 3,472 |
| 2025-12-02 | 30 Oct 2025 | 21 | 3,472 |
| 2026-01-06 | 30 Oct 2025 | 21 | 3,472 |
| 2026-02-11 | 30 Oct 2025 | 21 | 3,472 |
| 2026-04-04 | 30 Oct 2025 | 21 | 3,472 |
| 2026-10-02 | 22 Sep 2026 | 18 | 2,704 |
Which sites block these agents in robots.txt is covered in our AI-Crawler & llms.txt Census; how sites signal AI preferences is in AI preferences in robots.txt, and Common Crawl’s view is in Common Crawl robots.txt denials.
Method
On 17 October 2026 we downloaded OpenAI’s chatgpt-user.json, gptbot.json and searchbot.json, Google’s five crawler and fetcher files from developers.google.com/static/crawling/ipranges/, and Perplexity’s two files. We merged overlapping prefixes before counting addresses. For history we asked the Wayback Machine for every distinct capture of each file and kept the first capture per month; “list dated” is the file’s own creationTime. Networks come from iptoasn.com’s IPv4 table, downloaded the same day. A separate script recomputes every number from the saved files.
Caveats
- A published list is address space a company says it may use, not a measure of traffic. A large block can sit mostly idle.
- Wayback captures are irregular, so some months are missing and a change may have happened earlier than the capture that shows it. One Googlebot capture from April 2026 was empty and is left out.
- Counts are IPv4 only. Google’s files also list IPv6 prefixes, which we count but don’t size.
- iptoasn.com reflects routing as seen in its data on the day; “not routed” means no announcement there, not that the addresses are unused.
Download the data
- Live lists (CSV): every list with URL, date, prefix and address counts, and networks.
- Wayback history (CSV): prefix and address counts for each monthly capture of six lists.
Cite as: ProxyPicker, “AI Fetcher IP Ranges”, 17 October 2026, https://proxypicker.com/ai-fetcher-ip-ranges/. Data: OpenAI, Google, Perplexity, Internet Archive, iptoasn.com.
Related: AI-Crawler & llms.txt Census · AI preferences in robots.txt · Common Crawl robots.txt denials.
Press & data requests: email [email protected] for interviews, the raw data behind this page, or corrections. See also our contact page.