AI Content Strategy in 2026: Why Volume Stopped Working

AI made content cheap, and in doing so made generic content worthless. About half of new articles on the web are now written by AI, and AI answers in search resolve the simple questions those articles target. A content strategy built on publishing more is now competing with both. The strategy that works in 2026 publishes less, and puts its effort into what a machine cannot produce: original evidence, real experience and a clear point of view.

The supply shock

Content marketing used to be limited by production capacity. Not anymore. Graphite, an SEO firm, analysed 43,000 English-language articles sampled from Common Crawl and found that AI-generated articles overtook human-written ones in November 2024. The share has hovered around half since then.

Graphite’s explanation for the plateau is telling: AI-generated articles tend to perform poorly in search. Publishers tried mass production, and much of it simply did not get seen.

The demand shock

At the same time, search itself changed. When Google shows an AI summary, people click less. Pew Research Center tracked the real browsing of 900 US adults in March 2025: users clicked a traditional result on 8% of visits with an AI summary, against 15% without one. See our analysis of AI Overviews and click loss.

Put the two together. More content is chasing fewer clicks, and the content most easily replaced by an AI answer is exactly the generic, summarising kind AI tools produce best.

What Google is asking for

Google has been consistent. Its February 2023 guidance said it rewards quality “however it is produced” and treats automation used mainly to manipulate rankings as spam. Its May 2025 guidance on AI search goes further: “Focus on making unique, non-commodity content.”

Google also says something useful for strategy: clicks from results pages with AI Overviews tend to be “higher quality”, with visitors more likely to spend time on the site. Fewer clicks, but from people who wanted more than the summary. The job of content is to be worth that click.

From volume to signal: five shifts

1. Publish evidence, not summaries. Original surveys, first-party data, tests, benchmarks and case results are what AI answers cite and what readers click through for. One piece of real data can do more than twenty explainers.

2. Show experience. Reviews by people who actually used the product, practitioners describing how they solved a problem, photos and video from the field. Experience is hard to fake and easy to recognise.

3. Take a position. AI tends towards the average answer. A clear, argued view, even a contrarian one, gives readers a reason to choose you and journalists a reason to quote you.

4. Build things, not just pages. Calculators, templates, comparison tools and checklists earn repeat visits and links, and do not get replaced by a paragraph in a summary.

5. Be the source that AI cites. Clear structure, direct answers near the top, named authors and organisations, and sources for every claim make content easier to quote accurately. See the lexicon entry on LLMs.

Where AI fits in the strategy

None of this means avoiding AI. It means using it on the right side of the work: research summaries, outlines, variants, translation and editing support, with people owning the angle, the facts and the sign-off. Our AI content ops playbook sets out that workflow, and in the EU the human sign-off also matters for AI labelling rules.

Measuring what matters

If clicks fall while quality rises, page views become a poor scorecard. Track instead:

• conversions and sign-ups per visit, not just visits;
• citations and mentions in AI answers and in the press;
• links earned by original research;
• returning visitors to tools and resources.

Google’s own guidance says the same: measure the overall value of visits, not only the count.

The bottom line

The cost of producing content has collapsed; the cost of producing trusted content has not. Brands that keep scaling volume will fight over a shrinking pool of clicks with machines. Brands that invest in evidence, experience and judgment will become the sources those machines have to cite.