Title Tag Test: What Moved CTR

Title tag rewrites are the cheapest intervention in SEO — minutes per page — and the claims about them ("we doubled CTR!") deserve the same skepticism as everything in this genre. So: a controlled test on our own site. Thirty pages with high impressions and below-expectation CTR, titles rewritten per the formula kit; thirty matched controls untouched; eight weeks of Search Console data both sides. Here's what actually moved, by how much, and which rewrite patterns did the moving.

The selection and the rewrites

Test pool: pages with 1,000+ impressions/month whose CTR ran clearly below their position's own site-wide norm (each page benchmarked against our own position-CTR curve, not generic industry tables — the honest baseline). Positions ranged 3–12. The rewrites, one pattern per page, logged: specificity upgrades (vague promise → number/timeframe/outcome: 11 pages), front-loaded keyword + honest promise restructures (8), question-to-statement or statement-to-question inversions matched to intent (6), and dead-weight trims — years, redundant brand tails, filler words pushing value past the 60-char truncation (5). Meta descriptions untouched (one variable). Re-indexing requested; the clock started at each page's first recrawl.

The results at eight weeks

1. Test pages: CTR +31% relative (aggregate, impression-weighted: 2.6% → 3.4%); controls: +2% — seasonal noise. Clicks to test pages rose ~29% on essentially flat impressions.
2. 22 of 30 improved, five flat, three declined — and the decliners taught the most: two had traded accuracy for punch (the click arrived, the promise-content mismatch presumably bounced it; one later drifted down a position — small sample, consistent direction), one lost a query variant its old title had matched verbatim. Rewrites can lose relevance while gaining appeal; check what the old title was ranking for first.
3. Pattern scoreboard: specificity upgrades won biggest (+40%+ relative CTR on average — numbers and concrete outcomes out-pull adjectives, again), trims performed astonishingly well per minute spent (+25% average for ten-second edits), restructures solid (+20%), inversions mixed — intent-matched question titles helped on question-shaped queries and did nothing elsewhere.
4. Position side-effects: mostly none — 26 pages held position ±1, which cleanly separates the CTR effect from ranking change (and keeps this test agnostic on the CTR-as-factor debate: the clicks are worth having either way, which was the point). The three risers sat at snippet-adjacent positions where we'd also matched question phrasing — suggestive, not proof.

The operating playbook this confirms

Quarterly, per the audit habit: pull the high-impression/low-CTR list against your own position-CTR curve, rewrite with the formula kit's specificity bias, log the date, read at four-plus weeks — expecting the portfolio shape we got: most pages up meaningfully, a few flat, a couple needing reversion (keep the old titles logged; reversion is a feature of the process, not a failure). At ~15 minutes per page returning ~30% more clicks on pages already earning impressions, nothing else in the series pays faster — and nothing else runs out faster either: the mine is the underperformers; once they're rewritten, the quarterly haul shrinks to new decay.

Frequently asked questions

Why benchmark against my own CTR curve instead of industry tables?

Because SERP furniture, brand strength and query mix make every site's position-CTR reality different — a "bad" 4% at position 3 in one niche is excellent in another drowning in ads. Your own curve isolates the actual underperformers; tables generate false positives both directions.

How long before a rewrite's effect is readable?

The title changes at recrawl (days), the data stabilises in 3–4 weeks of impressions — read at four to eight, never at one (early numbers are position-mix noise). Annotate the change date anywhere you report, per the honesty rules.

Should I A/B test titles properly instead?

True A/B isn't available (one title per URL per crawl), so time-series-with-controls is the honest instrument — this design, roughly. What matters more than methodology polish: the specificity bias, the accuracy floor (the decliners' lesson), and doing it at all — the mine sits unworked in most Search Console accounts while budgets chase costlier levers. The costlier lever that is worth it afterward: the authority that raises the positions your better titles get shown at (us).