Long-form vs short-form: what 1000 posts of data actually show

The 'long-form always wins' fight died back in 2023 and yet I still see agencies handing in 3000-word briefs for transactional queries. I pulled a dataset of 1000 posts from 12 clients across Ahrefs and GSC between Jan 2024 and Mar 2026, cross-referenced intent through manual labeling plus GPT-5, and the results contradict most of the SEO folklore. Short posts (under 800 words) win 41% of commercial queries. Long posts (3000+) dominate deep informational, but with a brutal diminishing-returns curve past 2200 words. What matters is not length: it is answer density per intent, and I only saw this clearly after applying Search intent: 4 types and how to map them on the SERP as the primary filter.

I split the sample by intent type using the four classic buckets (informational, navigational, commercial, transactional) and the pattern jumped out immediately. For 'best X for Y' queries, top-3 median was 1400 words, not the 3000 people keep parroting. For 'how to X' tutorials, median was 2100. For 'X vs Y' comparison queries, surprise: 1800, because users want a comparison table, not an essay. For pure transactional like 'buy X', median 600 words. Anyone writing 2500 words on a product page is sabotaging conversion and cannibalizing the blog, a problem I unpack in E-commerce on-page: PLP vs PDP without cannibalization. The signal is clean enough that it became an internal rule: every brief starts from intent, never from a word target.

Second finding was the 'long-form without substance' effect. I split long posts into two groups: high unique-entity density (measured via TextRazor API, normalized per 100 words) and low density. High-density posts averaged 4.8% CTR in positions 1-3. Low-density posts, same length and position, hit 2.1%. Same SERP, same slot, half the clicks. Google does not explicitly penalize filler, but users penalize it in scroll behavior. Headings rescue part of this when structured well, and that impact I cover in Headings H1-H6: the structure Google actually reads. Without it, long posts become SEO tombstones.

I also looked at content decay and the result scrambles the 'long lasts longer' narrative. Of 1000 posts, 312 lost more than 30% of traffic over 18 months. Length distribution between decayed shorts and longs was nearly identical. What differentiated was update cadence and answer depth versus whatever the competitor published after you. 900-word posts with quarterly updates held traffic better than 3500-word posts left alone. If you are on the 'publish and forget' team, you will see decay regardless of size. I covered the identification framework in Content decay: spotting the posts quietly losing traffic and the rewrite-vs-rebuild call in Rewrite or rebuild: making the call with SERP data, worth revisiting before kicking off any new brief.

There is a third factor nobody measures: editorial opportunity cost. A team of four producing eight 3000-word posts per month ships 24k words. The same team doing twenty 1200-word posts ships 24k words with 2.5x more sitemap entries, more interlinking surface, more title and meta experiments. We ran that swap on three clients in the last year. Average result: +34% organic clicks in six months with the same headcount. The multiplier came from the link mesh that became possible with more nodes in the graph, following the model in Smart interlinking: the internal authority map. Page volume worked as the lever, not word volume per page.

To close with something actionable: stop writing toward a length. Define intent first, pull the 8-12 real questions users ask (use People Also Ask, Reddit, Search Console filtered to non-brand queries), answer each at the minimum depth required, and stop. If the sum is 700 words, ship 700. If it is 2400, ship 2400. Add a TL;DR up top to compete for featured snippets, technique I broke down in Featured snippets: how to structure content for position zero. Measure CTR per position against the 2026 benchmark and rebuild anything sitting 30% below curve. Length is an output, not an input. Anyone inverting that order burns time, budget and ranking to a competitor who figured it out first.