The formula has two sides, and only one of them moves
An agency’s content economics is one line of arithmetic: what the client pays, minus what production costs. The price side gets renegotiated every quarter. The cost side barely moves, and most of it never reaches an invoice.
There is not much room in that line. Promethean Research puts the average digital agency at a 13% after-tax net margin in 2025, down from 14% in 2024 and below the industry’s long-run average of about 15%. At that level the hours nobody billed are not a rounding error on a piece of content. They are the difference between a piece that earns and one that does not.
Start with the part that does reach an invoice. The Editorial Freelancers Association surveyed more than 1,100 of its members between November 2025 and mid-January 2026, and its 2026 rate chart puts the median range for a work-for-hire blog post at 25 to 45 cents a word. For a 1,500-word post that is $375 to $675. That figure buys a draft. It does not buy a published article.
Behind the draft sit hours, priced the same way. Orbit Media’s 2026 blogging survey of 1,042 content marketers puts the average time to write a blog post at 3 hours and 20 minutes. Your freelancer has costed those hours into the quote. The hours on your side of the transaction are the ones nobody quotes:
- writing the brief and choosing the angle
- answering questions while the draft is open
- reading the draft against the brief
- sending revisions and reading the revision
- finding the source for the claim that arrived without one
- formatting, linking and loading into the CMS
- absorbing the client’s own round of comments
The invoice is the only cost of a piece that anyone writes down.
Every one of those is paid at your team’s loaded hourly cost, by someone senior enough to make the call. None of it appears next to the writer’s invoice, so the piece looks cheap right up until you count.
Meanwhile the client has read about AI too
Two asks arrive, usually in the same conversation. Lower the price, because AI writes for free now. Or hold the price and raise the volume, for the same reason. Both assume the invoice was the cost.
That assumption is not going away, and arguing with it head-on loses. The answer that holds is a cost structure that genuinely changed shape, which is a different conversation from a discount.
A cheaper writer, human or machine, does not fix the leak
Swap the freelancer for a generator and the vendor line drops. Watch where the work goes.
Ahrefs surveyed 879 marketers about how they use AI in content production. Only 4% publish primarily “pure” AI content. Nearly all the rest run some kind of review first, and manual review is the most common method, used by 80%. Generation got cheaper. Everything after it did not, and it lands on the same manager who used to review freelancers.
That review is not skippable. A generated draft states a guessed figure in the same calm sentence as a checked one, so reading for flow never catches it, and someone has to open the source for every number. That is senior time, and it is billed to nobody.
A cheaper draft moves the cost from an invoice you forecast to hours you never track.
So the invoice fell and nothing else did. Cost moved from an external line you could forecast to internal hours you do not track. Content agency margins get thinner in exactly this way, quietly, while the vendor column of the P&L looks better than last quarter.
The cost does not disappear, it changes shape
| What you are looking at | Freelancer model | AI-writer model | Pipeline model |
|---|---|---|---|
| Where the spend sits | External invoice per piece | A low subscription or per-word price | A price per finished, checked article |
| Who spends the hours | The writer bills for them, your manager adds briefing, review and revisions | Your team, on editing, verification and rewriting | Your team, on the angle and one final read |
| What the manager does | Reviews a draft, then reviews it again | Repairs a draft that reads finished | Approves or returns a piece that arrives checked |
| Research and fact-checking | Trusted to the writer, verified unevenly | Falls to whoever edits | Runs as its own stage, with an output you can read |
| Shape of the cycle | Brief, draft, revise, revise | Prompt, edit, edit again | Brief, run, review once |
| What scaling adds | More writers and more coordination | More editing hours per person | More runs at the same overhead |
No numbers in that table, on purpose. Cheapness is not the claim being made here: the first two models move the same work between a vendor and your payroll, while only the third removes some of it.
Where the margin actually comes back
Not from paying a person less. From cutting the number of manual stages between a topic and a publish-ready piece.
Look at the freelance cycle honestly. The manager writes a brief, waits, reads, marks up, waits, reads again, then does the last mile personally because a third round costs more than fixing it. That loop, not the writing, is where the internal hours go. Speeding up any single step inside the loop changes very little, because the loop itself is the cost.
A pipeline attacks the loop. Research runs against live search data before a draft exists, so the angle is set on evidence rather than on the manager’s memory. Claims are checked one by one against the page they came from, so the piece arrives with its sources attached instead of with a question mark on every number. What reaches the manager is a piece to approve or return, not raw material to repair.
The margin comes back when the review loop stops, not when one step in it gets faster.
That is why the gain is not a percentage shaved off one operation. A stage either passes or the piece waits, so the review-revise-review cycle stops happening rather than getting faster. Judgment stays with your team: which topics are worth the client’s money, what they will not say in public, when a piece is wrong in a way no checklist catches. Those are the hours worth paying senior people for.
What to sell when “we use AI” means nothing
It already means nothing. In the 16th annual content marketing survey by Content Marketing Institute and MarketingProfs, fielded in mid-2025 across 1,015 B2B marketers, 95% said their organizations use AI-powered applications. A capability that close to universal cannot be a differentiator. Your client has heard it from everyone else on the shortlist.
What is not universal is evidence. We measured 531 articles ranking in Google’s organic top 20 for 40 B2B content-marketing queries (US, September 2026). Across them the median page carries 0.9 statistics per 1,000 words. Ranking pages are the competitive set, not a random sample of the web, which makes that thinness an opening rather than a complaint.
So sell the control points instead of the toolbox: research on live search data, a claim-by-claim fact-check with a record of what was checked against what, and a quality gate with its thresholds written down. Each of those is something you can show a buyer. Priced that way, the conversation moves off cost per word and onto what the client is actually buying, which is turnaround they can plan and claims they can defend. It also defends the margin, because it charges for the part of the work that costs you real money.
Where to start
Take the last five pieces you shipped and reconstruct them in hours, not invoices: briefing, questions, first read, revision rounds, the last mile someone quietly absorbed, client comments. Put the vendor cost next to that total. If the invoice turns out to be the smaller part, the gap tells you which stage to remove first.
Frequently asked
How do agency content production costs actually break down?
Into one visible line and several invisible ones. One line is the writer: the Editorial Freelancers Association’s 2026 rate chart, from a survey of more than 1,100 members, puts a work-for-hire blog post at 25 to 45 cents a word, or $375 to $675 for 1,500 words. Invisible are your own hours on briefing, mid-draft questions, review, revision rounds, sourcing and CMS work, priced at the loaded cost of whoever does them. Orbit Media’s 2026 survey of 1,042 content marketers puts the writing itself at 3 hours and 20 minutes, the part the freelancer already charged for.
Does switching to an AI writer improve agency margins?
It improves the vendor line, which is not the same thing. In the Ahrefs survey of 879 marketers, only 4% publish primarily “pure” AI content, and nearly all the rest review it first, most often by hand. The editing, verification and rewriting land on the manager who used to review freelancers, so the external saving reappears as internal hours that nobody bills.
What breaks first when scaling content production agency-side?
Coordination, before quality. Each additional writer adds a briefing thread, a review queue and a revision cycle, and those grow faster than output does. Models that scale are the ones where a piece passes fixed stages instead of circulating between people, because the overhead per additional piece stops rising with headcount.
Is "we use AI" still a differentiator?
No. Content Marketing Institute and MarketingProfs found that 95% of B2B marketers say their organizations use AI-powered applications, in a survey of 1,015 B2B marketers fielded in mid-2025. Everyone on the shortlist says it. What separates proposals now is what happens between the prompt and the publish button, and whether you can show it.
Sources
6 sources
Methodology: every figure above was first checked against the page it is cited from on September 8, 2026, and every source was fetched again on October 20, 2026, when the Orbit Media figures moved to its 2026 survey and the writer rates moved to the Editorial Freelancers Association’s 2026 chart, replacing a rates report whose survey dated to 2021. Items 1 to 5 are the publishers’ own research: an agency profitability benchmark, two production and rate surveys, one study with a published sample, and one industry survey. Item 6 is our own measurement, and its rules are stated with it.
- 01Promethean Research. “How Profitable are Digital Agencies?”, published April 19, 2026. https://prometheanresearch.com/how-profitable-are-digital-agencies/.Supports: the average digital agency earning a 13% after-tax net margin in 2025, down from 14% in 2024 and below the long-run average of about 15% since 2015. Figures are the averages from the publisher’s own agency surveys.
- 02Editorial Freelancers Association. “2026 Rate Chart” and “Editorial Rates”, rate chart published March 2026. https://www.the-efa.org/rates/, chart PDF https://www.the-efa.org/wp-content/uploads/2026/03/2026-Rate-Chart.pdf.Supports: the member survey run from November 2025 through mid-January 2026 with over 1,100 respondents, and the median range of 25 to 45 cents a word for work-for-hire blog posts (the $375 to $675 for 1,500 words is our arithmetic). Median ranges reported by the association’s members; rates vary with expertise and turnaround.
- 03Crestodina, A. “Blogging Statistics 2026: What 1,042 Content Marketers Told Us About What Works.” Orbit Media Studios, September 8, 2026. https://www.orbitmedia.com/blog/blogging-statistics/.Supports: the average time to write a blog post being 3 hours and 20 minutes, from 1,042 respondents. The fielding period is not stated on the page.
Show all 6 sourcesShow fewer
- 04Law, R. “Marketers Using AI Publish 42% More Content.” Ahrefs, June 11, 2025. https://ahrefs.com/blog/marketers-using-ai-publish-more-content/.Supports: the survey of 879 marketers, only 4% publishing primarily “pure” AI content, and manual review being the most common review method.
- 05Content Marketing Institute and MarketingProfs. “B2B Content and Marketing Trends: Insights for 2026”, published October 8, 2025. https://contentmarketinginstitute.com/b2b-research/b2b-content-marketing-trends-research.Supports: 95% of B2B marketers saying their organizations use AI-powered applications, the 16th annual survey fielded between June 24 and August 14, 2025, and the base of 1,015 B2B marketers.
- 06Avoid Content corpus measurement, September 2026. The instrument we ran over the corpus covered 531 articles ranking in Google organic top 20 for 40 B2B content-marketing queries (US, English), SERP pulled September 7, 2026.Supports: median 0.9 statistics per 1,000 words. Dataset not yet public; thresholds stated inline.