Which Pages AI Assistants Actually Read: 14 Days of Server Logs

Trusted by 810,000+ Techpresso subscribers·Editorial policy·How we make money

Between 3 and 16 November 2026, AI assistants opened a dupple.com guide 4,084 times while answering a person. "Is X free" and cost pages drew the most reads per guide (7.04 and 6.23), then best-of lists (4.28). "What is X" pages drew the fewest (0.88). Claude-User made 65.4% of the fetches and ChatGPT-User 30.6%.

Dupple publishes this study and sells placements in its guides. See AI visibility placements. The data is Dupple's own server logs and nothing else.

Summary

Summary: Which Pages AI Assistants Actually Read: 14 Days of Server Logs. Last updated November 2026.
Question Answer (3 to 16 November 2026)
How many times did an assistant open a guide for a user? 4,084 fetches across the 14 days
Which assistant? Claude-User 2,672, ChatGPT-User 1,250, DuckAssistBot 158, MistralAI-User 4, Perplexity-User 0
Which formats get read most per guide? "Is X free" 7.04, cost and pricing 6.23, best-of lists 4.28
Which get read least? What is X 0.88, how-to 1.63, alternatives 1.96
How many of the 1,075 live guides were read at least once? 824 (76.7%); the median guide was read 2 times
How concentrated is it? The top 10% of guides (107) took 39.2% of fetches
Do assistants ask for the Markdown version? 592 Claude-User fetches (22.2% of its 2,672) requested a .md URL
Are read pages also crawled by AI search indexers? 958 of 1,088 read pages (88.1%, pre-merge)

How we measured

  • Source. The nginx access logs of dupple.com. One site only.
  • Window. 3 November 2026 00:00 UTC to 16 November 2026 23:59 UTC, 14 days.
  • What counts. GET requests answered 200 or 304 on /learn/<slug> pages, which are Dupple's guides, and on the Markdown version of a guide (/learn/<slug>.md). The catalog counted 1,075 live guides.
  • User-triggered fetches. User agents Claude-User, ChatGPT-User, Perplexity-User, MistralAI-User and DuckAssistBot. Per their vendors' documentation, the first three open a page when a person's request calls for it: OpenAI says ChatGPT-User visits pages when a user's action in ChatGPT calls for it, Anthropic says Claude-User is used when people ask Claude questions and Claude fetches a page to answer, and Perplexity says Perplexity-User may visit a page to help answer a question. MistralAI-User and DuckAssistBot are counted in the same class here, and no vendor document for them is cited.
  • AI search indexers. OAI-SearchBot, PerplexityBot and Claude-SearchBot. OpenAI and Perplexity describe OAI-SearchBot and PerplexityBot as the bots that surface sites in their search results, and Anthropic describes Claude-SearchBot as improving Claude's search results.
  • Training and other crawlers. GPTBot, ClaudeBot, CCBot, Bytespider, meta-externalagent and Amazonbot.
  • Search engines. Googlebot, bingbot and Applebot.
  • Format. Each guide is assigned a format from the words in its slug: checked in this order, first match wins: "vs" is X vs Y, "alternative" is Alternatives to X, "cost", "pricing", "price" or a "how-much" start is Cost and pricing, a "best-" start is a best-of list, "is-...-free" is Is X free, "what-is" or "what-are" is What is X, a "how-to" start is How to. Everything else is Other.
  • Merging. A .md fetch is counted on its guide. Figures in this page are merged unless marked otherwise. Four items below (the table and FAQ split, the crawler .md counts, the indexer overlap and the top pages) use the pre-merge counts, where a guide and its .md URL count as two pages, and are labelled.
  • Reads per guide. User-triggered fetches divided by the number of live guides in that group.

Two caveats on the logs. User agents are self-declared, and they were not checked against the vendors' published IP ranges, so a script could copy one. And Perplexity-User logged no fetch in the window, while PerplexityBot, the indexer, logged 1,789. The logs do not say why, and this study does not guess.

Finding 1: Claude-User and ChatGPT-User made 96% of the fetches

Finding 1: Claude-User and ChatGPT-User made 96% of the fetches: Which Pages AI Assistants Actually Read: 14 Days of Server Logs. Last updated November 2026.
User agent Fetches Share of 4,084
Claude-User 2,672 65.4%
ChatGPT-User 1,250 30.6%
DuckAssistBot 158 3.9%
MistralAI-User 4 0.1%
Perplexity-User 0 0%

Per day, fetches ran from 547 on 4 November to 118 on 15 November, and weekends were not uniformly lower: 7 and 8 November logged 292 and 251, while 14 and 15 November logged 253 and 118.

Finding 2: cost and free-plan questions get the most reads per guide

Finding 2: cost and free-plan questions get the most reads per guide: Which Pages AI Assistants Actually Read: 14 Days of Server Logs. Last updated November 2026.
Format Live guides Fetches Guides read at least once Reads per guide Share of guides read
Is X free 28 197 27 7.04 96.4%
Cost and pricing 48 299 33 6.23 68.8%
Best X (ranked list) 626 2,680 536 4.28 85.6%
Other 104 384 62 3.69 59.6%
X vs Y 52 167 42 3.21 80.8%
Alternatives to X 49 96 37 1.96 75.5%
How to 142 232 76 1.63 53.5%
What is X 26 23 11 0.88 42.3%

Best-of lists are 58.2% of live guides (626 of 1,075) and took 65.7% of fetches (2,680 of 4,078 on live guides). The two highest rows are small groups (28 and 48 guides), so one popular guide moves them.

Among the 25 most-read single pages (counted before merging, so a guide and its .md URL are separate), the HTML pages included newsletter advertising cost with 40, newsletter CPM explained with 34, best POS systems with 32, is CapCut free with 28, best product analytics tools with 27, ChatGPT Plus vs Pro and Superhuman pricing with 23 each, and Google Workspace pricing with 21.

Finding 3: most guides get read, but a tenth take 39% of the reads

  • 824 of 1,075 live guides (76.7%) were opened at least once. 251 (23.3%) were not opened at all.
  • The median guide was opened 2 times in 14 days.
  • The top 10% of guides (107) took 39.2% of all fetches on live guides.

For an advertiser the useful reading is the spread: reads are not confined to a handful of pages, but the pages that draw them draw many.

Finding 4: updated best-of lists were read more, and so were longer ones

Within best-of lists only, which removes the format effect:

Finding 4: updated best-of lists were read more, and so were longer ones: Which Pages AI Assistants Actually Read: 14 Days of Server Logs. Last updated November 2026.
Group Live guides Fetches Reads per guide
Updated in the last 30 days 353 1,794 5.08
Updated 31 to 90 days ago 138 404 2.93
Updated 91 to 180 days ago 130 467 3.59
Updated more than 180 days ago 5 15 3.00
2,000 words or more 580 2,579 4.45
Under 2,000 words 46 101 2.20

Lists updated in the last 30 days averaged 5.08 reads. All older lists combined (273 guides, 886 fetches) averaged 3.25, so recently updated lists averaged 1.56 times as many. Lists of 2,000 words or more averaged 4.45 against 2.20 for shorter ones, but only 46 lists are under 2,000 words.

Length does not help everywhere. Among cost and pricing guides, the 16 under 2,000 words averaged 6.88 reads and the 32 at 2,000 words or more averaged 5.91. Among how-to guides, longer ones did better: 2.02 against 1.42.

Both splits are correlations, and the data does not show why recent or longer lists were read more. Among best-of lists the 31 to 90 day group (2.93) read less than the 91 to 180 day group (3.59), so the pattern is not a straight line.

Finding 5: guides with a table or an FAQ were read more, but the gap shrinks within one format

Pre-merge counts, all 1,075 guides:

Finding 5: guides with a table or an FAQ were read more, but the gap shrinks within one format: Which Pages AI Assistants Actually Read: 14 Days of Server Logs. Last updated November 2026.
Group Guides Fetches Reads per guide Share read at least once
With a table 998 3,346 3.35 75.8%
Without a table 77 141 1.83 58.4%
With an FAQ 993 3,284 3.31 76.1%
Without an FAQ 82 203 2.48 54.9%

These are correlations. Nearly every ranked list has a table (625 of 626), and ranked lists are read more per guide than how-to pages. Where a format has both kinds, the gap is smaller (merged counts): how-to guides with a table averaged 1.75 reads (96 guides) and without one 1.39 (46 guides). This data does not isolate table or FAQ as a lever.

Finding 6: one in five Claude-User fetches asked for the Markdown version

592 of Claude-User's 2,672 fetches (22.2%) requested the .md URL of a guide, which is 14.5% of all 4,084 user-triggered fetches. No .md request by ChatGPT-User, DuckAssistBot or MistralAI-User appears in the logs.

Crawlers asked for Markdown too (all logged fetches on /learn, by crawler):

Finding 6: one in five Claude-User fetches asked for the Markdown version: Which Pages AI Assistants Actually Read: 14 Days of Server Logs. Last updated November 2026.
Crawler .md fetches All logged fetches Share
GPTBot 74 86 86.0%
Applebot 1,235 1,635 75.5%
ClaudeBot 723 1,008 71.7%
Amazonbot 792 1,157 68.5%
meta-externalagent 985 2,756 35.7%
OAI-SearchBot 680 4,835 14.1%
bingbot 480 1,800 26.7%
PerplexityBot 2 1,789 0.1%
Googlebot 4 344 1.2%

Finding 7: read pages are almost all known to an AI search indexer

Before merging, 958 of the 1,088 pages an assistant opened (88.1%) were also fetched by OAI-SearchBot or PerplexityBot in the same window. This does not show that being indexed causes a read, only that the two overlap heavily.

Finding 8: crawler volumes by bot

Finding 8: crawler volumes by bot: Which Pages AI Assistants Actually Read: 14 Days of Server Logs. Last updated November 2026.
Class Fetches Bots (fetches)
AI search indexers 6,624 OAI-SearchBot 4,835; PerplexityBot 1,789; Claude-SearchBot none logged
Training and other crawlers 5,717 meta-externalagent 2,756; Amazonbot 1,157; ClaudeBot 1,008; Bytespider 633; GPTBot 86; CCBot 77
User-triggered 4,084 see Finding 1
Search engines 3,779 bingbot 1,800; Applebot 1,635; Googlebot 344

The four classes add up to 20,204 logged fetches, and AI search indexers were the largest class at 32.8%, with OAI-SearchBot alone making 73.0% of that class.

What this means if you want to be cited

All three points are tied to the data above, not to a claim about how any assistant ranks sources.

  1. Answer the cost and free-plan question on the page. "Is X free" and cost guides had the highest reads per guide (7.04 and 6.23), and 96.4% of "is X free" guides were read. This data shows the format is read, not that its content wins. See how to get cited in ChatGPT answers.
  2. Keep lists updated. Best-of lists updated in the last 30 days averaged 5.08 reads against 3.25 for older ones, a correlation only. Our guide to GEO tools covers what to track.
  3. Publish a Markdown version. Claude-User asked for it in 22.2% of its fetches. If you run a site, serve a .md twin of each guide with the canonical page linked.

On paid placements, read do sponsored listings improve AI visibility. For how Dupple's own audience figures are measured, see the newsletter engagement benchmarks.

Limits

  • One site, 14 days. The data covers dupple.com guides from 3 to 16 November 2026, and other sites, topics and weeks can differ.
  • Self-declared user agents. They were not checked against the vendors' published IP ranges, so some fetches may not come from the named assistant.
  • A fetch is not a citation. An assistant opening a page does not show that it quoted or linked the page in its answer.
  • Fetches are not people. One user question can trigger several fetches, so the counts are not a count of questions.
  • Slug-based formats. The format groups come from slug words, so a page can be misfiled, and the Other group is mixed.
  • Small groups. "Is X free" has 28 guides, cost and pricing 48, and "What is X" 26.
  • Perplexity-User logged no fetch while PerplexityBot logged 1,789. The logs do not explain it.
  • No causes. Freshness, length, table and FAQ are correlations, not tested effects.

FAQ

Which pages do AI assistants read the most?

On dupple.com, "is X free" and cost and pricing guides: 7.04 and 6.23 fetches per guide in 14 days, against 4.28 for best-of lists and 0.88 for "what is" pages.

Which AI assistant fetched the most pages?

Claude-User, with 2,672 of 4,084 user-triggered fetches (65.4%), then ChatGPT-User with 1,250 (30.6%).

Do AI assistants read the Markdown version of a page?

Claude-User did: 592 of its 2,672 fetches (22.2%) were for .md URLs. ChatGPT-User logged none.

Does updating a page help AI visibility?

In this data, best-of lists updated in the last 30 days averaged 5.08 reads and older ones 3.25. That is a correlation on one site, not proof of cause.

Do tables and FAQs make a page more likely to be read?

Guides with them were read more, but nearly every list has a table, and the gap shrinks within one format. Within how-to guides the gap was 1.75 against 1.39.

Is a bot visit the same as being cited?

No. A user-triggered fetch means an assistant opened the page. These logs do not show whether the page was quoted.

Can I buy a placement in these guides?

Yes. Dupple sells placements in its guides, see AI visibility placements.

Placements in Dupple's guides: dupple.com/advertise/ai-visibility.

Some offers on this page may be paid placements or contain affiliate links.

Advertise to 970K+ tech professionals

Dupple is a distribution platform for tech and finance products. Reach developers, founders, executives and marketers through sponsorships in Techpresso, Cyberpresso, Finpresso, Devshot, MarketingShot and AIpresso, lead generation, and listings on Toolradar. Copy review included, rates in our media kit.

See ad formats Request the media kit