From Sanskriti Khandelwal | Product & Market Analysis
Content Marketing in 2026: Posts Got 4.7x Cheaper, Clicks Went the Other Way
On this page
The average cost of producing a blog post fell from $611 to $131 once teams moved drafting to AI. Budgets did not fall with it. The same money now buys roughly five times the pages, at every competitor at once. Meanwhile the share of US Google searches that produce any click at all fell 22.9% between 2024 and 2026. That is not a content quality problem. It is a supply glut meeting a shrinking market.
Key takeaways
- Cost per post fell 4.7 times and budgets stayed flat. Ahrefs surveyed 879 marketers and found $131 per AI-assisted post against $611 per human-written post, with monthly content spend of $2,475 and $2,442 respectively. The saving was converted into volume, not banked.
- The click supply shrank while the page supply expanded. SparkToro's analysis of Similarweb clickstream data put US zero-click searches at 68.01% in early 2026, up from 60.45% in 2024. Searches producing at least one click fell 22.9% over those two years.
- Google's rule is about value at volume, not about AI. Its spam policy defines scaled content abuse as many pages generated "for the primary purpose of manipulating search rankings", and applies that test "no matter how it's created".
- The flood landed hardest on definitional content. In Ahrefs' sample of 55.8 million AI Overviews, 97.70% sat on informational queries and only 19.97% on branded ones. The cheapest content to automate was also the content most likely to be answered without a click.
Why did publishing more stop working? Because the cost of a page fell about 4.7 times while the number of searches that send a click fell 22.9%. Supply expanded and demand contracted inside the same two years. More output now buys a smaller share of a smaller pool, and every competitor made the identical move at the identical moment.
The first thing the flood changed was the cost of supply
Most commentary treats the content glut as a quality story. It is more usefully read as a price story, because a price change explains the behaviour and a quality complaint does not.
When the marginal cost of producing a unit falls and the budget stays constant, output rises. That describes what B2B publishing did between 2024 and 2026 with uncomfortable precision.
The measured numbers, not the impression
Ahrefs surveyed 879 marketers and published the result in June 2025. The average self-reported cost of a blog post was $131 for teams using AI against $611 for teams not using it, a difference of 4.7 times.
The distribution matters more than the average here. 87% of AI users reported spending $0 to $100 per post, against 39% of teams not using AI. At the top end, 11% of non-AI users spent over $1,000 a post while only 2% of AI users did.
Those are self-reported figures from a vendor survey, so treat them as directional rather than audited. No competing dataset argues that drafting got more expensive.
The budget did not move, so the volume did
Here is the finding that should have reorganised more content plans than it did. Monthly content spend was $2,475 for AI users and $2,442 for teams not using AI. The two budgets are effectively identical.
So the saving did not come back as saving. Run the arithmetic on those averages and a $2,475 monthly budget buys about 4 posts at $611 and about 19 posts at $131.
Ahrefs also found that only 38% of respondents reported saving money on writers, while 45% reported no writer saving at all.
My view is that this was the most consequential decision most content teams made in the period, and almost nobody made it deliberately. It was the default outcome of a flat budget meeting a cheaper input.
The denominator almost nobody prices into a content plan
Publishing plans are usually built on an expected traffic number per post. That number has a denominator, and the denominator has been moving against publishers for three years.
Start with the base rate. Ahrefs analysed roughly 14 billion pages in its Content Explorer index in December 2023 and found 96.55% of them received zero traffic from Google. Another 1.94% received between one and ten monthly visits.
That figure predates the current flood, which is precisely why it is useful. Publishing has always had a brutal failure rate. The flood did not create the problem, it raised the number of entrants competing inside an already thin 3.45%.
Clicks are the scarce input now
SparkToro analysed Similarweb clickstream data covering US Google searches from January to April 2026. 68.01% of searches ended without a click, against 60.45% in 2024.
Expressed the other way, the share of searches generating at least one click fell 9.51 percentage points, a relative decline of 22.9%. The share of searches that led to another Google search instead rose 7.2 percentage points.
Only 0.34% of searches transitioned into AI Mode during the study period. The click loss is happening inside ordinary search results, not inside a separate product most people have not adopted.
Pew measured the mechanism at the user level
The Pew Research Center ran the cleanest available test of why this is happening. It tracked 900 US adults who installed a browser monitoring app, capturing 68,879 unique Google searches during March 2025, of which 12,593 produced an AI summary.
Users clicked a traditional search result on 8% of visits where an AI summary appeared, against 15% where it did not. Clicks on the links inside the summary itself accounted for 1% of visits.
Session behaviour shifted too. 26% of pages carrying a summary ended the browsing session, against 16% of pages without one. Pew also found that 18% of all sampled searches produced a summary, and 88% of those summaries cited three or more sources.
Read those two studies together and the picture is consistent. Citation is not the scarce thing, since summaries name several sources each. The click is the scarce thing. That gap between being cited and being visited is now the central problem, and it is examined in the evidence on what actually drives AI citations.
Why ten times the output produced nothing like ten times the result
Put the two sides together and the disappointment stops being mysterious. It becomes predictable, and it was predictable in advance.
Hold your share of the market constant for a moment. The pool of clicking searches shrinks to 77% of its former size while the number of competing pages rises 4.7 times. Expected clicks per published post then land near one sixth of where they started. You would need to publish six times as much simply to stand still.
That is arithmetic on two published averages, not a measurement of your site. Output tripled across the category and traffic did not follow it.
| What changed | Direction | Figure | Source |
|---|---|---|---|
| Cost to produce one blog post | Down sharply | $611 to $131, a 4.7x fall | Ahrefs, 879 marketers, June 2025 |
| Monthly content budget | Flat | $2,475 AI users, $2,442 non-AI users | Ahrefs, June 2025 |
| Posts the same budget buys | Up about 5x | About 4 to about 19 per month | Arithmetic on the two rows above |
| US searches ending without a click | Up | 60.45% to 68.01% | SparkToro and Similarweb, 2026 |
| Searches producing at least one click | Down | 9.51 point fall, a 22.9% relative decline | SparkToro and Similarweb, 2026 |
| Indexed pages getting zero Google traffic | Already very high | 96.55% of 14 billion pages | Ahrefs, December 2023 |
The third row is derived, not surveyed. It divides the reported budget by the reported cost per post and assumes a team spends its whole budget on articles, which most do not. Treat it as the shape of the shift rather than a production plan.
No single team was wrong to publish more, which is what makes this hard to argue with internally. If your cost per page drops and every competitor's cost drops identically, holding output flat surrenders share to whoever does not hold.
The trap is that the payoff depends on what everyone else does, and everyone else faced the same maths on the same day. This is the structure of a commons problem. It resolves the way commons problems always resolve, badly and quickly.
Compute pricing produced the same pattern one layer down, where per-token prices fell while total bills rose, a dynamic covered in the analysis of why cheaper tokens produced larger invoices. Cheap inputs do not reliably produce cheap outcomes. They produce more consumption.
The flood did not spread evenly across query types
Aggregate numbers hide the part you can act on. The click loss is concentrated exactly where automated content is easiest to produce.
Ahrefs analysed 55.8 million AI Overviews across 590 million keywords in its index in May 2025. 97.70% of AI Overviews appeared on informational queries and 19.97% on branded ones. Local searches accounted for 6.85%.
The commercial split is just as sharp. 71.67% of searches showing an AI Overview had no cost-per-click data attached, meaning advertisers were not bidding on them. The queries being absorbed are largely the ones nobody was paying to reach.
Glossary pages, "what is" explainers and beginner overviews were the first things most teams pointed a model at. They are also the exact pages an AI summary replaces most completely. A 67 word median summary, which is what Pew measured, fully answers a definitional query. It cannot fully answer a question about which of two vendors handles your specific edge case.
| Query type | Share of AI Overviews | What that implies for the content type |
|---|---|---|
| Informational | 97.70% | Highest exposure. Cheapest to automate, most likely to be answered in place. |
| Non-branded | 80.03% | High exposure. Generic category terms are where every competitor published. |
| Branded | 19.97% | Low exposure. Demand for your name still resolves to your site. |
| Local | 6.85% | Lowest exposure. Physical proximity is not summarisable. |
| Queries with no CPC data | 71.67% | Most absorbed queries were never commercially contested to begin with. |
Categories are not mutually exclusive, so the column does not sum to 100%.
Google is not refereeing the fight you think it is
Most frustrated teams I speak to believe Google penalises AI writing. That belief is checkable against the published policy, and it is wrong in a way that costs real time.
Google's spam policies define scaled content abuse as "when many pages are generated for the primary purpose of manipulating search rankings and not helping users". The policy then adds that it targets "large amounts of unoriginal content that provides little or no value to users, no matter how it's created".
That last clause carries the whole rule. Production method is not the violation. Volume without value is the violation, and it always was. A hand-written page of nothing and a generated page of nothing are the same object under this policy.
So the enforcement risk is not really an AI risk. It is a cadence risk. It becomes real the moment your publishing rate exceeds the editorial capacity that would catch a worthless page before it ships.
Be careful with the penalty figures circulating
Search for enforcement data and you will find confident claims that penalised sites lost 50% to 80% of their traffic, usually attached to a named core update. I went looking for the underlying sample behind those numbers and did not find one.
Every version I opened traced back to a vendor blog with no stated methodology, no site count and no time window. Six restatements of an unsourced figure is not corroboration, it is one unsourced figure. Those percentages are not in this post.
What survived the flood
The useful question is not what to stop making. It is which assets still cost something to produce, because production cost is now the only reliable barrier left standing.
Anything a competitor can reproduce for $131 provides no defensible position, because they will reproduce it. In the surviving categories, the expensive input is not the writing.
First-party numbers are the clearest case. A benchmark drawn from your own customer base cannot be generated by anyone else. The cost sits in instrumentation and a quarter of patient collection, and no model shortens that.
Hands-on comparison is the second case. Testing two products against a real workload requires access, time, and a willingness to publish an unflattering finding.
Named practitioner accounts are the third. A person describing a decision they made, with the outcome and the part they got wrong, is not reproducible by anyone who did not live through it.
| Asset | Why volume did not commoditise it | What it actually costs |
|---|---|---|
| Benchmark from your own data | Nobody else holds the sample | Instrumentation, plus a collection period you cannot compress |
| Hands-on product comparison | Requires access and real usage, not description | Licences, test time, and the risk of an unflattering finding |
| Named practitioner account | Tied to one person's actual decisions | Internal time from someone whose time is expensive |
| Branded demand | Only 19.97% of AI Overviews sit on branded queries | Years, and spend on channels that are not search |
| Generic explainer at volume | It did not survive | $131, which is the entire problem |
The adoption gap is the actual opening
The 16th annual B2B research from the Content Marketing Institute and MarketingProfs, fielded across more than 1,000 B2B marketers, found 90% now use AI to produce content while fewer than 40% report an actual performance improvement.
Their stated 2026 investment plans explain part of that. 45% expect to increase spending on AI-powered marketing tools. Investment in people, meaning salaries, training and development, ranked last at 9%.
I think that ratio is inverted. When 90% of a market holds the same tool, the tool is not the variable. The variable is whoever still employs someone capable of producing a number nobody else has. The same adoption-without-return pattern is showing up in the macro data, and it is unpicked in the piece on the AI productivity paradox.
96% of B2B marketers say they create thought leadership content, and in many programmes fewer than 5% of employees contribute to it. The gap between that ambition and that staffing is where differentiation was supposed to come from.
Where this argument is weakest
This post makes a strong claim from datasets that were never designed to be combined. Here is what a careful reader should push back on.
The datasets do not line up in time or geography
The cost survey is from June 2025, the zero-traffic study from December 2023, the click data from early 2026 and US-only, and the AI Overviews sample from May 2025. Multiplying figures drawn from different years and different populations is a rhetorical device rather than a measurement. That is why the relevant chart is labelled illustrative.
Three of the seven sources behind this post also come from a single vendor whose product is sold to people worried about exactly this problem. That is a real conflict and you should weight the figures accordingly.
Not every team converted the saving into volume
The flat-budget finding is an average across 879 respondents. Some teams took the money out, some held output flat and raised quality, and some never adopted AI at all. The five-times-more-pages conclusion describes the market in aggregate, not any specific competitor set.
The 96.55% zero-traffic figure also needs a caveat that gets dropped whenever it is quoted. It counts every indexed page, including many never intended to rank at all, and Ahrefs notes that its index skews toward the higher quality end of the web.
The part I cannot settle
SparkToro explicitly declined to isolate how much of the zero-click rise AI Overviews caused. The correlation is strong and the mechanism Pew measured is plausible, but search behaviour was already drifting toward zero-click before summaries existed.
Anyone telling you the exact share is guessing. The direction of travel is not in doubt, and it alone justifies changing a budget.
Reallocating a content budget that stopped paying
The instinct when traffic falls is to publish more, because that is the lever with the shortest feedback loop. On the numbers above, that lever now works against you.
The alternative is unglamorous and takes a quarter to show anything. Cut publishing volume by half, hold the budget flat, and spend the freed money on one asset per quarter with an input cost a competitor cannot match in an afternoon.
Measure it on assisted pipeline rather than on sessions, because sessions are the metric the market has just destroyed. Traffic arriving from AI surfaces also behaves differently from ordinary organic traffic, a difference quantified in the analysis of how AI referral traffic converts. That changes what a smaller number is worth.
One warning before you retire old pages. Being cited is not the same as being visited. A page earning no clicks may still be feeding the answers your buyers read, a distinction examined in the comparison of ChatGPT citations against Google rankings. Check citation presence before deleting anything. The overlap between those two surfaces is weaker than most teams assume, as set out in the work on the rank and citation overlap collapse.
Frequently asked questions
Why is my content not ranking even though I publish more?
Because your competitors increased output at the same time and the available clicks shrank. Ahrefs found the cost of a blog post fell from $611 to $131 while content budgets stayed flat, so the same money buys roughly five times the pages. Meanwhile SparkToro measured US searches producing at least one click falling 22.9% between 2024 and 2026. The result is more supply chasing less demand.
Is AI generated content bad for SEO in 2026?
Google's spam policies do not prohibit AI writing. They prohibit scaled content abuse, defined as many pages generated primarily to manipulate rankings rather than to help users, and the policy applies that test "no matter how it's created". The real risk is publishing at a cadence your editorial review cannot cover. A worthless page is a worthless page whoever or whatever wrote it.
How much does it cost to produce a blog post with AI?
Ahrefs surveyed 879 marketers in 2025 and found an average of $131 per post for teams using AI, against $611 for teams not using it. 87% of AI users reported spending $0 to $100 per post, compared with 39% of non-AI users. Average monthly spend on AI tools was $188, with 64% of companies under $500 a month.
What is scaled content abuse and how do I avoid it?
Google defines it as generating many pages primarily to manipulate search rankings rather than to help users, including using generative tools to produce pages that add no value. Avoiding it is a question of editorial capacity rather than tooling. If nobody with judgement reads a page before it publishes, your publishing rate has exceeded your review rate and the policy is the least of the problem.
What content still gets clicks in 2026?
Branded and local queries, mostly. Only 19.97% of AI Overviews in Ahrefs' 55.8 million sample sat on branded queries and 6.85% on local ones, against 97.70% on informational queries. Content built on first-party data, hands-on product testing or a named practitioner's actual decisions also holds up, because it cannot be reproduced by anyone lacking the underlying access.
Should I publish less content and make it better?
Publishing less only helps if the freed budget buys something a competitor cannot copy cheaply. Halving output and making the remaining pages slightly more polished changes nothing, because polish is now free. Halving output to fund one original benchmark a quarter changes the input cost of your content, which is the only barrier the flood did not erode.
Where to start this week
Two tasks, both finishable before Friday, both producing a number you can act on.
First, divide your quarterly content spend by the number of pieces you shipped, then divide your organic pipeline contribution by that same figure. If cost per piece fell over the last two years while pipeline per piece fell faster, you have confirmed the trade you made without meaning to make it.
Second, list every asset you published last quarter and mark the ones a competitor could not reproduce for $131. If that list comes back empty, you now know precisely what next quarter's budget is for, and the number to beat is sitting on your own invoice.
Related analysis
The measurement side of this problem is covered separately in the citation evidence for GEO against SEO and in the breakdown of where measurable AI return has actually appeared.
References
- Ahrefs, AI Content Is 4.7x Cheaper Than Human Content, 18 June 2025. Survey of 879 marketers. Used for all cost per post, budget and writer saving figures.
- Ahrefs, 96.55% of Content Gets No Traffic From Google, 1 December 2023. Roughly 14 billion indexed pages. Used for the zero-traffic base rate.
- Search Engine Land, Google zero-click searches reach 68% in early 2026, 2026. Reporting on the SparkToro study of Similarweb US clickstream data, January to April 2026. Used for all zero-click figures. The underlying SparkToro publication is the stronger citation and should replace this one.
- Pew Research Center, Google users are less likely to click on links when an AI summary appears, 22 July 2025. 900 US adults, 68,879 searches. Used for click rate, session end and summary prevalence figures.
- Ahrefs, Insights From 55.8M AI Overviews Across 590M Searches, 19 May 2025. Used for all query type shares and the CPC data figure.
- Content Marketing Institute, New B2B research finds winning marketing teams are building fundamentals, 2026. With MarketingProfs and Storyblok, more than 1,000 B2B marketers. Used for adoption, performance and investment figures.
- Google Search Central, Spam policies for Google web search. Used for the scaled content abuse definition and quoted text.
Three of the seven references come from Ahrefs, a company that sells tools to the audience most alarmed by these findings. Those figures are consistent with the independent Pew and Similarweb data used alongside them, which is the only reason they are quoted without a stronger caveat.
Related reading