Measuring AI Content ROI

Why content ROI is hard to measure

Content rarely produces a sale on first visit. It earns rankings over months, is read by people who buy later through another channel, and is used by sales teams in conversations analytics never sees. Measured badly, it looks worthless for six months and then gets credited with everything.

AI changes one side of the equation sharply: the cost of producing a piece falls. It does not change how slowly search traffic builds or how indirectly content sells. So the measurement problem is the same as ever, with one new temptation: publishing so much that nobody checks whether any of it works.

Step 1: Cost each piece fully

Illustrative example with hypothetical round figures. An AI-assisted article takes:

• • 1 hour of research and briefing at a loaded £35 an hour: £35
• • AI-assisted drafting and editing, 2 hours: £70
• • Fact checking and review by a second person, 1 hour: £35
• • Tools, shared across the month's output: £10

Full cost per article: £35 + £70 + £35 + £10 = £150.

Leave out review and you would record £115 and overstate the return by the difference. Review is also what keeps the content in line with Google's guidance, which focuses on the quality and helpfulness of content rather than how it is produced, and treats using automation to generate content primarily to manipulate rankings as spam.

Step 2: Track by publication cohort

Group articles by the month they were published and follow each cohort for twelve months. Illustrative assumptions for one article in its first year:

• • Months 1 to 3: negligible traffic while it is indexed and ranks
• • Months 4 to 12: 200 organic visits a month
• • Visit to lead conversion: 1%
• • Lead to customer: 10%
• • Gross profit per customer: £1,500

Leads per month from month 4: 200 × 1% = 2.

Customers per month: 2 × 10% = 0.2.

Gross profit per month: 0.2 × £1,500 = £300.

Months of earning in year one: months 4 to 12 is 9 months. 9 × £300 = £2,700.

Year-one return on the £150 article: £2,700 minus £150 = £2,550.

That looks excellent, and it is the reason to be careful. Most articles in a cohort do not perform like the average; a few earn most of the traffic. So calculate the cohort, not the single article. If a cohort of 10 articles cost £1,500 and only 2 of them reach 200 visits a month while the rest earn nothing, the cohort's year-one gross profit is 2 × £2,700 = £5,400, still well above cost. If none reach it, the cohort earns nothing, and the low cost per piece did not help.

Step 3: Count what content assists, not only what it closes

Last click undercounts content, because the reader usually returns later by search, email or direct visit. Two practical fixes:

• • Analytics: in GA4, look at conversion paths and compare the reporting attribution model you use; data-driven attribution gives content that starts or assists journeys more credit than last click.
• • CRM: record the first content page a lead visited and ask new customers what they read. Sales teams can tag articles they sent during deals.

Report assisted value separately from last-click value, and say which is which.

Step 4: Use traffic value carefully

A common shortcut values organic traffic at what the same clicks would have cost in paid search. It is a proxy for reach, not a measure of return.

AfM's founder-owned brand Harmonance publishes an example of the method: its case record (October 2024 to March 2025, 152 days measured) gives an "Equivalent Paid Media Value" of "GBP 5,488", which is its 10,976 organic clicks × an assumed £0.50 a click. Change the assumed cost per click to £2 and the same clicks become £21,952; nothing about the site's performance changed, only the assumption. That figure describes what the traffic would have cost to buy, not revenue earned; the record gives no revenue figure. Harmonance is owned by AfM's founder and is not a client result. It is cited here only to show the method and its limit.

Content ROI methods and their limits

Method	What it shows	Limit
Equivalent paid traffic value	What the organic clicks would have cost in paid search	Not revenue; depends entirely on the assumed cost per click
Last-click conversions	Leads or sales where content was the final touch	Undercounts content that starts or assists journeys
Assisted conversions and data-driven attribution	Content's share of multi-touch journeys	Only sees tracked digital touchpoints
CRM first-touch and sales tagging	Content that began or supported real deals	Relies on disciplined data capture
Cohort payback	When a month's content has earned back its full cost	Needs twelve months of patience and tracking

Step 5: Decide what to keep producing

After two or three cohorts have run six months or more, compare them by topic type and format. Keep producing what reaches the payback line; update pieces that rank but convert poorly; stop producing the topics that never earn impressions. That is where AI's low production cost helps most: updating and improving existing pages is cheap, and often pays better than adding new ones.

For a comparison of production models (manual, AI-assisted with editorial review, and autonomous), see Content Marketing ROI: Autonomous vs Manual Production. For when content returns should be expected relative to other channels, see AI Marketing Investment Timeline.

FAQ

How long should I wait before judging a piece of content?

Judge cohorts, not single pieces, and give search content most of a year. Early signals such as impressions and ranking positions can be read from the first few months.

Does Google penalise AI-generated content?

Google's guidance focuses on whether content is helpful and reliable, not on how it is made. Using automation mainly to manipulate rankings breaches its spam policies.

Should I count social shares or time on page as ROI?

They are useful diagnostics for whether content is read. They are not return; tie them to leads, assisted conversions or sales conversations before reporting value.

Is updating old content worth measuring separately?

Yes. Track updated pages as their own cohort from the update date. Updates often cost little and can recover rankings faster than new pages build them.