Content marketing hit a 12-year performance low, and AI adoption does not explain it

Content marketing performance has fallen to its lowest point in twelve years, and the survey that found the drop says AI adoption is not what caused it. Orbit Media's 2026 blogging survey asked 1,042 content marketers whether their blog delivers results. Only 13.9% put it in the strongest category, the lowest share the survey has recorded since it began in 2014.

For more than a decade that figure sat between 20% and 30%. This year it fell six points below the previous low. A second number travelled just as far: 18% of respondents now say they do not know whether their blog delivers results at all, against a range of 9% to 14% across the previous decade.

AI made the work faster and changed nothing else

Orbit found 92.4% of respondents now use AI somewhere in their blogging process, and none of the individual use cases it tested showed a relationship with strong results. The average post takes three hours and twenty minutes to write, down from a peak of four hours and ten minutes in 2022, and runs 1,312 words, down from 1,427 in 2023. Orbit calculates that the typical respondent spends roughly fifty fewer hours a year writing than in 2022.

Orbit's own reading is that the decline comes from what marketers stopped doing. Six habits tied to strong performance all thinned out over the past year: working with outside experts, paying to promote posts, researching keywords, watching analytics with any discipline, running original studies, and sending drafts through a real editor. The habits tied to weak performance all grew. Expert collaboration is the sharpest predictor in the whole set, beating the benchmark by 2.6 times, and it has emptied out faster than anything else on the list, practised by 25% of respondents in 2017 and 7% now. Running original studies improves the odds by about half, and each year fewer teams bother.

They might look like extra steps. But they're often where the greatest value enters the work.

That is Ann Handley of MarketingProfs, quoted in Orbit's report, on the steps being cut. Orbit tracks eight such practices. Respondents who kept six or more of them going reported strong results at close to triple the rate of the field. Not one of the 1,042 ran all eight.

Why this reads differently at a software company

Orbit, a web design and development agency that has run this survey annually since 2014, polled content marketers across industries, not B2B software specifically, so read the headline figure as a direction rather than a benchmark for your team. The programme it describes as the failing default is unmistakably the standard SaaS blog: text-heavy how-to posts, few visuals, no named human attached, pushed out by email and a company social account.

The abandoned practices also share a property that bites harder in software than elsewhere. Each of them needs a second person. An expert to interview, an editor to argue with, a buyer to place the promotion, an analyst to build the dataset. AI removed the cheapest link in that chain, the first draft. On a lot of software teams the expensive links went too, because the headcount performing them was the line item the drafting tool was bought to justify cutting.

The measurement finding lands hardest. Orbit argues traffic has become a poor proxy in a market where buyers ask an assistant for a recommendation. If the assistant names you, the visit arrives as a branded or direct session at a high conversion rate. If it does not, nothing happens and there is nothing to count. Respondents who named pipeline figures such as qualified leads, closed deals and revenue did better on the results question than those watching sessions.

The agency Kalungi points the same way after working through raw Search Console exports for a set of B2B SaaS sites. Once branded and login queries were stripped out, click-through on discovery terms from the top three positions sat at or under 1.2% for most of them, with the full spread running from 0.2% to 12%. Kalungi sells B2B SaaS marketing services and says plainly that this is a pattern to check against your own data, not a benchmark. Search Engine Land, which also covered the Orbit survey, is owned by Semrush and sells keyword tooling.

What to change before the next budget cycle

• Count how many of the eight practices your programme still runs. Fewer than six explains more about your results than your AI stack does.
• Rebuild original research first. Your product generates data nobody else holds, and a number that did not previously exist is the one asset a language model cannot manufacture.
• Put a named human editor in front of publication, and named experts inside the pieces.
• Move the reported metric off sessions and onto pipeline now, because the people who cannot say whether the blog works will lose the budget argument.

The honest caveat is that self-reported performance and workable attribution collapsed in the same window. Some of the lost confidence is a measurement failure rather than a marketing failure. That distinction offers less comfort than it should. A programme whose results you cannot demonstrate is already in trouble, whatever is happening underneath it.