The arithmetic is fine. The inputs stopped covering the return
Return minus cost, divided by cost. That formula hasn’t aged badly. What aged badly is the habit of filling in the return term from a traffic report.
SparkToro read Similarweb’s desktop and mobile panel for January to April 2026 and reported that “in the first four months of 2026, a whopping 68.01% of Google searches ended without a click.” Inside Google’s AI Mode the share is higher still. Semrush took clickstream data for almost 69 million Google Search sessions between May 1 and July 5, 2025, kept the US desktop ones, and found that “92-94% of AI Mode searches were zero-click searches.”
Both are third-party estimates, and Google disputes the method behind that kind of study. In August 2025 Liz Reid, who runs Google Search, wrote that “total organic click volume from Google Search to websites has been relatively stable year-over-year,” and put reports of a collapse down to “flawed methodologies, isolated examples, or traffic changes that occurred prior to the roll out of AI features in Search.” Google published no click data alongside the post.
For a budget defense the disagreement matters less than the direction. Part of the demand your content creates now happens where your analytics has no instrument, and a content marketing ROI calculation built on sessions records that part as zero.
The return that never reaches your analytics
Picture a page that answers one narrow buying question well. An engine uses it to assemble an answer, names your brand while doing so, and the reader never opens a tab. Your dashboard logs nothing. That reader may still arrive three weeks later through a branded search, already knowing who you are.
An answer that names you is exposure your dashboard records as nothing.
The tempting shortcut is to treat rankings as a stand-in for that exposure. They’re a poor one. Ahrefs put 15,000 long-tail queries to Google and Bing, asked the same questions of AI assistants, and reported in August 2025 that “only 12% of links cited by ChatGPT, Gemini, and Copilot appear in Google’s top 10 results” for the same prompt. The engines aren’t selecting a copy of the ranking, so the new input can’t be derived from the old report.
Freshness is one visible difference between the two lists. A separate Ahrefs analysis of 16.975 million cited URLs across ChatGPT, Perplexity, Gemini, Copilot, AI Overviews and organic Google results found cited URLs to be, on average, 25.7% “fresher” than the URLs sitting in organic results. That fact belongs in the cost model rather than the visibility slide.
Two terms the calculation is missing
On the cost side: the hours, and then the upkeep
Production cost usually enters the model as an invoice, when it enters at all. Hours are the steadier unit. Orbit Media’s 2026 blogging survey of 1,042 content marketers puts the average time to write a post at “3 hours, 20 minutes.” That covers writing. Briefing, research, fact-checking, editing, visuals and publishing sit outside it, and in most B2B teams those hours belong to salaried people whose time never reaches the content budget line.
Dollar rates depend on market and seniority, and one published rate chart is not your invoice; our piece on agency margins walks through a 2026 rate survey if you need a reference point. Take your own last ten invoices, add internal hours at loaded cost, and take that number into the budget meeting.
One line is almost always missing from it. Seer Interactive dated 7,683 cited pages carrying 47,097 citations across ChatGPT, Gemini and Perplexity between March and June 2026. Of the 4,124 pages where both dates were readable, 72% had been updated within the past year, but only 42% had been published within it. Most citations were landing on older pages that someone had kept current. A page that earns citations is an asset with a maintenance schedule, so a per-piece cost that ignores the updates it will need over the next two years is understated.
On the value side: citation share, and what buyers type into the form
Two inputs, both cheap. The first is citation share. Fix a list of fifteen to twenty questions your buyers actually ask, phrased the way they would type them, run it through the assistants your market uses, and log which domains get named. Repeat monthly with the same list. It produces one number, the share of your priority questions where an engine names you, and one list, the questions where it names somebody else. That list is next quarter’s brief.
The other input is self-reported attribution. Put a “how did you first hear about us” field on high-intent forms and read the free text. “Heard about you from ChatGPT” is a weak signal on its own, easy to over-read, and it’s the only signal a reader who never clicked will ever hand you. Count it, keep it in its own column, and resist folding it into attributed revenue.
| Term | Old formula | Full formula |
|---|---|---|
| Cost | The freelance or agency invoice | Invoice, plus internal hours, plus the updates the page will need |
| Reach | Organic sessions | Sessions, plus the share of tracked questions where an engine cites the page |
| Conversion | Requests attributed in analytics | Attributed requests, plus self-reported answers naming an assistant, kept separate |
| Time frame | The reporting quarter | The working life of the page, including the updates that keep it cited |
| A zero in the report | The page failed | The page failed, or the instrument does not reach where it worked |
A worked example with hypothetical numbers
Two articles, one team, one quarter. Every figure in this example is hypothetical, invented to show the arithmetic. None of it is measurement from a live account.
Article A came from a freelancer and pulled more clicks. Article B came out of a staged pipeline at a fraction of the cost and gets cited on nearly half of the twenty questions the team tracks.
| Hypothetical example, illustrative numbers | Article A | Article B |
|---|---|---|
| Production cost, invoice plus internal hours | $900 | $300 |
| Organic sessions, first six months | 620 | 410 |
| Demo requests attributed in analytics | 8 | 5 |
| Tracked buyer questions where an engine cites the page, of 20 | 0 | 9 |
| Form answers naming an AI assistant, same period | 0 | 4 |
| Cost per attributed request | $113 | $60 |
| Cost per request including self-reported answers | $113 | $33 |
_Hypothetical example, illustrative numbers. Not data from any account._
Read the old report and Article A wins. Eight requests against five, more traffic, and cost that sits inside a retainer nobody itemizes. Commission more like A.
The old report picked the winner before anyone looked at the missing columns.
Add the two missing columns and the ranking flips. B produced demand at a third of the cost per request, and nine of the twenty tracked questions now return an answer with your brand in it, which is exposure that arrives whether or not anyone clicks. Neither column is precise. Both are closer to the truth than a zero.
This isn’t an argument that pipelines always win. Move the citation row to A and A wins on every line. It is an argument about which numbers decided the call, and here they were the two the old report never had.
What the full picture changes
Content ROI now needs a real cost on one side and a measurement of AI visibility on the other. Teams counting clicks and rankings alone aren’t working from wrong data. They are working from a frame that ends before part of the value is delivered, which produces confident decisions with a piece missing. The same missing piece shows up in the quarterly report, where content metrics that report a business result replace an activity log.
Start smaller than a project. Rebuild the cost of your last ten articles with internal hours included, then run twenty buyer questions through two assistants and write down who gets named. That’s an afternoon of work, and it tells you whether the pieces your report calls weak are working somewhere it can’t see.
Frequently asked
What goes into a content marketing ROI calculation now?
The same fraction, with two terms added. On the cost side, the invoice plus internal hours plus the updates a page needs to stay useful. Seer Interactive found that of cited pages with readable dates, 72% had been updated within the past year but only 42% published within it, so upkeep is part of what a cited page costs. On the return side, add the share of tracked buyer questions where an engine cites you, counted separately from attributed revenue.
Why do zero-click searches break the classic formula?
Because the return term is read off a traffic report. SparkToro measured Similarweb’s panel for January to April 2026 and found “68.01% of Google searches ended without a click,” and Semrush reported “92-94% of AI Mode searches were zero-click searches” from US desktop sessions in its 2025 clickstream study. Google disputes the methodology behind third-party estimates of this kind. Even so, enough answers are delivered above the link that a sessions-only return is incomplete.
Can rankings stand in for AI visibility?
Not closely enough to use. Ahrefs put 15,000 long-tail queries to Google, Bing and several assistants in 2025 and found “only 12% of links cited by ChatGPT, Gemini, and Copilot appear in Google’s top 10 results” for the same prompt. Rank and citation answer different questions, so tracking one doesn’t give you the other.
How much does an article cost to produce?
There’s no rate we would quote as a benchmark. Orbit Media’s 2026 survey of 1,042 content marketers put writing time at “3 hours, 20 minutes” per post, which is the labor unit and not the bill. Add briefing, research, fact-checking, editing and publishing hours at your loaded rates, plus the updates the page will need. That figure is the denominator.
What is the cheapest way to start measuring citations?
A list and a calendar. Take fifteen to twenty questions your buyers ask, run them monthly through the assistants your market uses, and record which domains are named. Keep the list stable so the months compare. Answers vary between runs, so read the result as a rate across the list rather than reacting to one miss.
Sources
7 sources
Methodology: every figure and quotation above was checked against the page it is cited from on October 27, 2026, and each quoted sentence appears on that page. All seven items are the publisher’s own research or statement, not a restatement by a third party.
- 01Fishkin, R. “In 2026, Less than One Third of Google Searches Still Send a Click.” SparkToro, June 9, 2026. https://sparktoro.com/blog/in-2026-less-than-one-third-of-google-searches-still-send-a-click/.Supports: 68.01% of Google searches ended without a click in January to April 2026, measured on Similarweb’s desktop and mobile web panel (US).
- 02Harsel, L. “Google AI Mode’s Early Adoption and SEO Impact.” Semrush, July 30, 2025. https://www.semrush.com/blog/google-ai-mode-seo-impact/.Supports: 92-94% of AI Mode searches were zero-click, from clickstream data for almost 69 million Google Search sessions between May 1 and July 5, 2025, limited to US desktop sessions.
- 03Reid, L. “AI in Search is driving more queries and higher quality clicks.” Google, August 6, 2025. https://blog.google/products-and-platforms/products/search/ai-search-driving-more-queries-higher-quality-clicks/.Supports: Google’s position that organic click volume is stable year-over-year and that third-party traffic studies rest on flawed methodologies.
Show all 7 sourcesShow fewer
- 04Linehan, L. “Only 12% of AI Cited URLs Rank in Google’s Top 10 for the Original Prompt.” Ahrefs, August 11, 2025. https://ahrefs.com/blog/ai-search-overlap/.Supports: 12% overlap between AI citations and Google’s top 10, from 15,000 long-tail queries put to Google, Bing and four assistants.
- 05Law, R. “New Study: AI Assistants Prefer to Cite ‘Fresher’ Content (17 Million Citations Analyzed).” Ahrefs, July 28, 2025. https://ahrefs.com/blog/do-ai-assistants-prefer-to-cite-fresh-content/.Supports: cited URLs averaging 25.7% “fresher” than organic results, across 16.975 million cited URLs.
- 06Vasquez, S. “Study: Content Recency’s Impact on AI Visibility in 2026.” Seer Interactive, July 24, 2026. https://www.seerinteractive.com/insights/study-content-recencys-impact-on-ai-visibility-in-2026.Supports: 7,683 dated pages carrying 47,097 citations across ChatGPT, Gemini and Perplexity, March to June 2026; of the 4,124 pages with both dates readable, 72% updated and 42% published within the past year.
- 07Crestodina, 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: average writing time for a post of 3 hours, 20 minutes, from a survey of 1,042 content marketers.