Content teams have never produced more, faster, for less certain return. The share of marketers reporting strong results from blogging just hit an all-time low of 13.9% — in the same year 92.4% of them adopted AI.
This page collects what is actually measured, with the sample, the method and the date attached to every figure. It also does something no competing round-up does: where four credible studies give four different answers to “how much of the web is AI-written”, it prints all four and explains why they differ.
Key Takeaways
- Only 13.9% of marketers report strong results from blogging, down from 26% in 2022 and the lowest in twelve years of surveys.
- AI improved productivity for 87% of marketers and content performance for 39% — the defining gap of the year.
- Roughly half of new online articles are primarily AI-written, and that share has been flat since early 2025.
- Google is not penalising AI content as a class — but heavily AI pages index at 40.4% versus 49.3% and earn 2–3x fewer impressions.
- How-to content keeps ~30% of its peak traffic a year later; research content keeps under 10%. Format decides the decay curve.
- 94% of articles earn zero external links and 1.3% of articles generate 75% of all social shares.
The Headline Numbers
Sources: Orbit Media Studios (survey, n=1,042, 8 September 2026) · Graphite (measured, 55,400 URLs, May 2026) · Ahrefs (measured, ~150,000 pages, 27 July 2026) · Content Marketing Institute (survey, n=1,015, 8 October 2025).
Read together, those four numbers describe the central problem of content creation in 2026: the industry has never produced content faster, and has never been less sure it is working.
How Much Published Content Is Actually AI-Written?
You will see this quoted as 10%, as 17%, as 50% and as 74%. All four numbers are real. They measure four different populations with four different detectors at four different thresholds, and stacking them into a trend line — as most round-ups do — is simply wrong.
| Figure | What population it measures | Source and method |
|---|---|---|
| 10% | All webpages sampled, July 2026 — including everything published before ChatGPT existed | Pew Research Center Data Labs, 490,000 pages from Common Crawl, Open Pangram detector |
| Over one third | Pages published after November 2022 only | Same Pew study — this is the figure that reconciles it with Graphite |
| 17.31% | Pages ranking in Google’s top 20 for 500 informational keywords | Originality.ai, ~10,000 sites per wave, to September 2025 |
| 49.9% | New articles of 100+ words, majority AI by volume | Graphite, 55,400 random URLs, three detectors averaged |
| 74% | New pages containing any detected AI content | Ahrefs — a much looser test, which is why it is highest |
The more interesting finding sits underneath the headline: Graphite measured AI articles reaching 35.9% within twelve months of ChatGPT’s launch and 48% within two years — and then plateauing at roughly 50% since Q1 2025. The flood levelled off eighteen months ago. Almost nobody reports that part.
Pew’s domain breakdown is worth keeping in view too: roughly 10% of .com pages show AI-authorship signs, against 4.6% of .org and about 1% of .edu and .gov.
Why You Should Distrust Any Single Detection Number
Every study above publishes limitations that the round-ups quoting them omit. In the publishers’ own words:
- Detectors work in aggregate, not per document. Pew states its conclusions are valid only across “very large collections of texts” and that models “sometimes misclassify individual documents that were written by humans as including signs of AI authorship, and vice versa”.
- Surface tells are not evidence. Pew notes explicitly that em dashes or Oxford commas “don’t necessarily mean a particular piece of writing was produced using AI”.
- Hybrid work is invisible. Graphite says outright that it did not evaluate AI-assisted content — human-edited AI drafts, which is how most professional content is now made. Berkeley’s Dan Klein told Axios the human/AI line is “fuzzier than the chart suggests”.
- Several detectors are sold by the companies publishing the studies. Originality.ai sells the detector behind its own figure; Ahrefs uses its own. Both disclose it; neither is disqualified by it, but it belongs next to the number.
- Ahrefs’ own caveat: “AI content detection is not perfect, and the way we detect AI content will be different from how Google does.”
Does AI Content Rank? The Measured Answer
This is the question everyone argues about from anecdote. Ahrefs answered it with data: one million pages from top-ten rankings across 100,000 SERPs, of which about 150,000 had enough text to test.
That is the whole effect. A 3.8-point gradient across the entire first page — not a cliff, not a penalty.
Source: Ahrefs, 27 July 2026. One million ranking pages, ~150,000 tested. Measured.
The Productivity–Performance Gap
The single most revealing pair of numbers in content marketing right now comes from asking marketers the same question two ways.
Source: Content Marketing Institute and MarketingProfs, published 8 October 2025. n=1,015 B2B marketers, fielded 24 June – 14 August 2025. Survey self-report.
Teams are producing more, faster, and fewer than half see it show up in results. CMI’s enterprise edition (n=296, January 2026) found the same shape: 84% productivity, 53% quality, 38% performance.
And measured behaviour corroborates the self-report. Orbit Media Studios found that AI users and non-users were equally likely to report strong results, with no individual AI use case beating the 14% benchmark — while non-users were more likely to report disappointing results. Whatever AI is doing, it is not moving the outcome.
Spending follows the productivity number rather than the performance one: 45% of B2B marketers will increase investment in AI tools in 2026, and 9% in people and training — the lowest-ranked item on the list. Meanwhile CMI’s 2026 Career and Salary Outlook (n=644) reports marketing layoffs up 30% year on year.
What Blogging Actually Delivers Now
Source: Orbit Media Studios, published 8 September 2026, n=1,042 content marketers. Self-reported. Orbit does not disclose fielding dates.
Posts are shorter, they take about 50 fewer hours per marketer per year than in 2022, nearly everyone uses AI — and the share reporting strong results is the lowest in the survey’s history, down from the 25–30% band that held from 2014 to 2019. A further 18% do not know whether their blog delivers results at all.
Two things in the same dataset do correlate with strong results:
- Publishing original research raises the likelihood of strong results by roughly 50% — and the share of marketers doing it is falling after nearly a decade of growth.
- Two to three visuals per post correlates most strongly with strong results. Most marketers add fewer.
How Fast Content Decays — and Why Format Decides
| Format | Median time to peak | Traffic lost within 3 months of peak | Retained later |
|---|---|---|---|
| How-to content (n=309) | 5 months | 61% | ~30% of peak at one year |
| Listicles (n=303) | 5 months | 82% | ~12% of peak at six months |
| Research and data (n=107) | 1 month | 89% | under 10% at one year |
| All content (n=719) | 4 months | ~76% | — |
Source: Fractl, published 14 September 2026. 719 pieces tracked by GA4 monthly views across three client portfolios (business finance, energy, home security), March 2024 – June 2026. Measured.
The pattern is counter-intuitive and useful: the format that earns the biggest spike decays fastest. Research content peaks within a month and is effectively gone within a year. How-to content peaks slowly and still holds about a third of its peak twelve months later. If you need compounding traffic rather than a launch spike, that is the trade-off, stated in data.
Word Count, Links and Shares
The most-quoted number in content briefing is “1,447 words”. It comes from an 11.8-million-result correlation study — whose authors found no direct relationship between word count and ranking position and wrote this caveat, which almost nobody reproduces:
The larger content study — 912 million blog posts, Backlinko and BuzzSumo, February 2019 — is seven years old but still the only dataset of its size, and it shows that optimal length depends entirely on what you want:
| Finding (2019 data, 912 million posts) | Figure |
|---|---|
| Extra links earned by long-form vs short articles | +77.2% |
| Extra shares for 1,000–2,000 words vs under 1,000 | +56.1%, with diminishing returns beyond 2,000 |
| Extra shares for list posts vs how-to posts | +218% |
| Extra referring domains for “why”/“what” posts and infographics vs video and how-to | +25.8% |
| Articles receiving zero external links | 94% |
| Share of all social shares generated by 1.3% of articles | 75% |
| Correlation between social shares and backlinks | Virtually none |
That last row is the most under-cited finding in the whole dataset. Shares and links are not proxies for each other, so a content programme measured on one is not measuring the other.
Content and AI Search Citations
Source: Ahrefs, 4 February 2026, and its April 2025 predecessor. 300,000 keywords, same method both times. Measured. Average CTR on AI Overview keywords fell to 0.016 from 0.073.
Both editions are cited here deliberately: the same study run twice, nine months apart, shows the impact roughly doubling. Articles that quote only one figure are hiding a time series.
What actually gets cited:
- 76.10% of pages cited in AI Overviews rank in the organic top 10, 9.5% rank 11–100, and 14.4% do not rank in the top 100 at all. Median position of the primary cited URL: 2. (Ahrefs, 1.9 million citations from 1 million AI Overviews, July 2025.)
- ChatGPT cites very differently: roughly 90% of webpages it cited ranked at position 21 or below in Google. (Semrush, July 2025 — but note this covered only 500+ digital marketing and SEO topics, one unusually AI-literate vertical.)
- Video dominates. The most-cited domains in AI Overviews are YouTube 23.3%, Wikipedia 18.4% and Google.com 16.4%. (Surfer, 36 million AI Overviews and 46 million citations, March–August 2025. Surfer publishes no methodology beyond the tracker name and date range.)
- AI-cited content is 25.7% fresher on average than pages ranking organically — the clearest argument available for a refresh cadence.
Video and Format Performance
Video is where the format data is strongest, and where the conventional wisdom is most wrong.
- Engagement falls as videos get longer — but action rises. Videos under five minutes hold attention best, yet lead-gen form click-through runs 53–55% on 5–30 minute videos and 67–75% on videos over an hour. Short form wins attention; long form wins intent.
- Webinar replays of 31–45 minutes drew more than double the engagement of replays under 30 minutes, and about a third of on-demand webinars still accumulate plays three months after the event.
- Optimal length is industry-dependent, not universal. Food, beverage and retail win under three minutes; finance, insurance and health see 30-minute-plus content outperform; software and tech peak at 3–30 minutes.
- LinkedIn is now B2B’s leading video channel at 81% usage, ahead of YouTube at 76%.
Source: Wistia, 22 April 2026. Survey of 900+ professionals plus platform analysis of 13+ million videos and 79 million hours of viewing. Mixed measured and self-report.
The widely quoted ROI figures come from a much smaller survey and should be labelled as such: 82% of video marketers say video has given them good ROI and 85% say it generated leads (Wyzowl, n=266, fielded late 2025, self-reported). A 266-person self-report is not a measurement of ROI — it is a measurement of belief.
What Content Costs
| Deliverable | Most common rate |
|---|---|
| 1,500-word blog post | $250–$399 |
| Whitepaper | $500–$999 |
| Email copywriting | $99–$249 |
| Ghostwriting | A 16–20% premium over bylined work |
Source: Peak Freelance, 6 November 2025. n=213 freelance writers. Self-reported; fielding dates not disclosed.
The distribution matters more than the median: more than half of freelance writers earn under $30,000 a year, while those earning over $100,000 typically charge at least $1,000 per blog post and a quarter of that group charge over $1,500 for a single 1,500-word article. And 42% of high-paying assignments arrive by word-of-mouth referral — more than social media and communities combined.
Content Statistics You Should Stop Citing
| The statistic | Where it actually comes from | Verdict |
|---|---|---|
| “Content marketing costs 62% less and generates 3x more leads” | a Demand Metric infographic, circulating by 2013 — no method, no sample, no fielding date, and gated | An infographic, not a study. Fourteen years old and still printed undated |
| “Companies that blog get 67% more leads” | Attribution contradicts itself across every citing source; no original publication exists | Untraceable. Also un-interpretable — companies that blog differ systematically from those that don’t |
| “1,447 words is the ideal blog post length” | The average length of pages already ranking, in a study that found no relationship between length and position | The caveat was removed in transit. Orbit’s measured 2026 average is 1,312 words and falling |
| “74% of web content is AI-generated” | Pages containing any detected AI content — a different question from “is AI-written” | Real, but not interchangeable with the 50% or 10% figures |
| “94% of content gets zero backlinks” | Backlinko and BuzzSumo, February 2019 | Sound — but seven years old and almost always printed undated |
| “52% of consumers disengage from suspected AI content” | Every trail ends at an AI-writing-tool vendor’s blog | No primary source exists. Same for “AI-assisted content performs 12% better” |
| Any undated HubSpot-attributed figure | HubSpot’s statistics hub carries no update date and no method for its own surveys | The largest single vector of circular citation in this niche |
| “CMI B2C benchmark 2026” | CMI’s 2026 cycle is B2B, Enterprise and Career/Salary only | There is no 2026 B2C report. Treat any such figure as misattributed |
What to Actually Do With This Data
Posts are shorter and faster to produce than in 2022, and the share of marketers reporting strong results is at an all-time low of 13.9%. Volume went up and outcomes went down; a dashboard built on publishing cadence will show the opposite of the truth.
87% report productivity gains and 39% report performance gains. Heavily AI-generated pages index at 40.4% versus 49.3% and earn two to three times fewer impressions. AI belongs in the middle of the process, not at the end of it.
How-to content holds roughly 30% of its peak a year later; listicles hold about 12% at six months; research content loses 89% within three months of peaking. Match the format to whether you want a spike or a compounding asset.
It raises the likelihood of strong results by about 50% in Orbit Media’s survey, and the number of teams doing it is falling — which is exactly when it is worth doing.
94% of articles earn zero external links and 1.3% of articles generate 75% of shares. A target that assumes the average article earns links is a target built on a number that does not describe any real article.
AI-cited content is 25.7% fresher than organically ranking content, and 76% of AI Overview citations come from the organic top ten. Keeping existing top-ten pages current is the cheapest AI-visibility work available.
Three of the most-quoted numbers in this industry are an undated 2012 infographic, an untraceable lead claim and a correlation whose authors warned against causal reading. Citing the source and the year protects you from all three.
Want Content That Actually Earns Rankings?
We plan, write and refresh content against measured search demand — and we show you the numbers behind every recommendation. Start with a free audit.
Get a Free Audit →Frequently Asked Questions
How much online content is AI-generated in 2026?
Has AI content growth slowed down?
Does Google penalise AI-generated content?
Can AI content rank in the top three?
Does using AI improve content results?
What is the average blog post length in 2026?
How quickly does content lose traffic?
How much do businesses pay for content?
What is the ROI of content marketing?
Which content statistics should I stop citing?
Sources
- Content Marketing Institute and MarketingProfs — B2B Content and Marketing Trends: Insights for 2026, 8 October 2025. n=1,015, fielded 24 June – 14 August 2025. Self-reported.
- CMI’s enterprise edition — 21 January 2026. n=296 enterprise marketers. Self-reported.
- CMI’s 2026 Career and Salary Outlook — 8 April 2026. n=644. Self-reported; fielding dates not disclosed.
- Orbit Media Studios — Annual Blogging Statistics Survey, 8 September 2026. n=1,042. Self-reported; fielding dates not disclosed.
- Graphite — AI Now Writes as Many Online Articles as Humans Do, May 2026. 55,400 URLs, three detectors averaged. Measured.
- Pew Research Center Data Labs — How Much of the Internet Is Written With AI?, 20 August 2026. 490,000 pages. Measured.
- Originality.ai — AI content in Google search results, updated 17 September 2025. ~10,000 sites per wave. Measured; vendor sells the detector used.
- Ahrefs — Google Doesn’t Punish AI Content, 27 July 2026. One million ranking pages, ~150,000 tested. Measured.
- Ahrefs — AI Overviews and click-through, 4 February 2026, with its 17 April 2025 predecessor. 300,000 keywords. Measured.
- Ahrefs — 21 July 2025. 1.9 million citations from 1 million AI Overviews. Measured.
- Fractl — Content Decay, 14 September 2026. 719 pieces, 28 months, three client portfolios. Measured; single-agency dataset.
- Backlinko and BuzzSumo — 912 million blog posts, 19 February 2019. Measured; not updated since.
- Backlinko and Ahrefs — 11.8 million search results, updated 14 April 2025. Measured correlation study.
- Surfer — AI Citation Report, March–August 2025. 36 million AI Overviews. Measured; no published methodology.
- Semrush — 21 July 2025. 500+ digital marketing and SEO topics. Measured, single vertical.
- Wistia — State of Video Report, 22 April 2026. 900+ professionals plus 13 million videos. Mixed.
- Wyzowl — Video Marketing Statistics 2026. n=266, fielded late 2025. Self-reported.
- Peak Freelance — Freelance Writing Rates, 6 November 2025. n=213. Self-reported.
Last reviewed 18 September 2026. Every figure above is labelled with its source and publication year; where two credible studies disagree, both are shown rather than the more convenient one.
Related statistics guides
- Google SEO Statistics 2026
- AI SEO Statistics 2026
- AI Overview Statistics 2026
- B2B SEO Statistics 2026
- Digital Marketing Growth Statistics 2026
- AI Search Statistics 2026
- Ecommerce SEO Statistics 2026
- Technical SEO Statistics 2026
- Email Marketing Effectiveness Statistics 2026
- Local Consumer Review Statistics 2026
- Local Search Ranking Factors 2026





