Skip to content
World clockEU--:--UK--:--USA--:--CN--:--PLDEFRIT中文EN

portal about AI and technologyevents · analysis · interviews · technical background

Search
LIVE
›

ChatGPT takes 95 percent of AI traffic to news sites. The rest barely registers

Data from BrightEdge shows that in August 2026 ChatGPT accounted for 95.1 percent of the traffic that AI assistants referred to publishers. Gemini, Perplexity and Claude split just under 5 percent between them. The promised diversification of AI channels turned out to be a myth.

Media & internetAnalysisRachel NwosuPublished: 23 September 20264 min readSources 3
ChatGPT takes 95 percent of AI traffic to news sites. The rest barely registers

Publishers who spent the past two years optimizing for generative search engines hoped for a scattered presence, a little bit from each assistant. The data BrightEdge collects paints a different picture. In August 2026, ChatGPT was the source of 95.1 percent of all traffic referred to content from tools built on language models.

Who splits the rest

Gemini accounted for 2.4 percent, Perplexity for 1.1 percent and Claude for 0.8 percent of that traffic. Referrals from ChatGPT rose 101 percent between January and August, and 36 percent in August alone. In absolute terms it is still a fraction of all traffic to news sites, but the pace of growth has no equivalent in any other channel.

The position of the competition shifts too: in August both Gemini and Claude recorded drops in referred traffic volume, by 15 and 17 percent respectively. Perplexity lost a similar share. Jim Yu of BrightEdge said the growth of ChatGPT alongside the weakening of its rivals points to consolidation, not to the start of the healthy competition publishers had expected.

Agents query more often than people

The report brings one more figure that should worry newsrooms. AI agents make about 88 queries for every 100 visits that users typing a query into a search engine used to generate. Each query is potentially a separate request for access to content, but only one of them ends with a human visit to the publisher's site. Traffic without people grows, traffic with people does not.

At the same time, the company 51Degrees found that over a year the traffic of so-called AI harvesters, the bots that collect content for models, grew 62.5 percent, from 8 to 13 percent of all requests. AmazonBot led with 4.1 percent of requests, ahead of Meta-ExternalAgent with 2.9 percent. BingBot, Anthropic, OpenAI, Perplexity and Gemini together account for fractions of a percent. The paradox is that the largest volume of referred traffic to news outlets comes from the entity whose own crawler collects content most aggressively.

Local outlets first in line

The scale of the problem varies with the type of publisher. Data from Semrush, cited by the technology press, showed that organic traffic from the United States to USA Today fell 18 percent between June 2025 and June 2026. According to Axios analyses, the smallest newsrooms feel it hardest, with traffic in the range of 10 000 visits per month. They have neither a team to test new formats nor bargaining power in licensing talks. For them the whole transformation comes down to one thing: fewer visits, lower revenue from programmatic ads and less capacity to keep the newsroom running.

Comments 0

Sources

3
  1. 01ChatGPT controls 95% of AI referral traffic to publishersEN
  2. 02Chatbot Traffic Rises 60% YoY, AI Harvesters Now 13% Of All RequestsEN
  3. 03Google pod presją wydawców. AI Overviews wywołuje kryzys w mediachPL

All figures and quotations in this text come from the sources listed below.

Content prepared by the editorial team with AI assistance.

Rachel Nwosu

Rachel Nwosu

AI, models and technology

Rachel Nwosu covers AI, models and technology for FLASH24, working from public model documentation, benchmark releases and repository histories rather than press summaries, and she skips announcements that arrive without reproducible numbers. She checks training-data claims against dataset cards and reruns reported metrics where code is available. She spends much of her week interviewing researchers and engineers, tracking model launch calendars, and comparing vendor benchmarks with independent evaluations. Outside the desk she runs 3D printers, restores old computers, and tests how models learn from internet junk. She does not publish benchmark figures she cannot trace to a source.

Newsroom →

Comments

0
  1. No comments yet — be the first.

Write a comment

Comments are public. We do not publish abuse, spam or advertising.