TL;DR
Generate 5 headline variants that each use a different framework and promise a different reader outcome, then sanity-check the top 1-2 against the live Google SERP before committing. The model writes the candidates; the SERP tells you which framework has white-space. Keep titles under ~580 pixels (roughly 55-60 characters) so Google does not truncate the benefit. Skip A/B testing unless a page sees 1,000+ visits per variant — below that you are reading noise.
The task
The article is done. You spent four hours on the body and four minutes on the headline, and the headline is the part 95% of readers will see. The first thing you wrote down was a competent-but-flat “How to write AI blog headlines that convert.” That headline has the keyword, the right intent, and nothing that would make anyone click it over the 12 other pages in the SERP with similar titles. You want 5 candidate variants spanning genuinely different angles, not five rephrasings of the same one, so you can lock in the one that gives you a real CTR advantage.
Which model to use (June 2026)
Headline generation is a short-form creative task, and the two strong picks split on personality. GPT-5.5 (ChatGPT default since April 2026; pick the Thinking mode in the model picker) is the better brainstormer: it reliably returns more genuinely distinct angles per pass, which is exactly what you want when forcing five different frameworks. Claude Opus 4.7 or Sonnet 4.6 writes a more natural, lower-edit headline with fewer AI tells, so it is the better finisher once you know which angle to keep. A practical split: generate the 5 variants with GPT-5.5 Thinking, then paste your favorite into Claude and ask it for three tighter rewrites. Gemini 3.1 Pro is the third option and the one to reach for if you also want it to fetch the live SERP in the same chat.
All three are free-tier capable for this task; you do not need a paid plan to generate headlines. See ChatGPT vs Claude vs Gemini for the full split.
Where AI helps, and where it does not
AI is excellent at applying established headline frameworks consistently: question, listicle, contrarian, how-to, before/after, status / identity, number-led. It is also good at keeping the keyword in the first half of the headline without sounding stuffed. Where AI fails: predicting search-intent shifts and ranking the variants against your specific SERP. The model does not know that for your target query, the top 3 SERP results are all “how-to” framings, meaning the contrarian or status angle has more white-space opportunity. Sanity-check the top 1-2 variants against the actual live SERP for your keyword before you commit. AI generates; the SERP decides.
A common failure mode: all 5 variants are stylistic siblings, same outcome promised with a different verb. That is one variant in five costumes. Force angle separation in the prompt: each variant must promise a different reader outcome, not just a different headline style.
What to feed the AI
- The article’s single biggest takeaway in one sentence (not the title — the takeaway)
- The primary target keyword exactly as readers search it
- The reader’s emotional state when searching this — frustrated, curious, comparison-shopping, ready-to-buy
- One example of a past headline of yours that performed well (tone anchor)
- 2-3 of the SERP’s existing top-ranking titles — so the model knows what the page is competing against
- The platform — Google search, Substack inbox, Twitter share preview, LinkedIn feed all favor different lengths and angles
- Length constraints — Google truncates on pixel width (about 580 px on desktop, roughly 55-60 characters), not raw character count, so wide letters like W and M eat the budget faster; aim under 90 characters for full social card previews and under 100 if the post might get featured in a newsletter
- Hard avoids — words you have banned for AI-tic reasons (“ultimate,” “unlock,” “elevate,” “delve,” “tapestry”)
Copy-ready prompt
Generate 5 headline variants for a blog post.
Constraints:
- Under 60 characters each, and avoid stacking wide letters (W, M, capitals) since Google truncates on pixel width (~580px), not character count.
- The target keyword appears in the first half of the headline as a natural noun phrase, not as an SEO anchor.
- Each variant uses a different framework AND promises a different reader outcome.
Article takeaway: {one sentence}
Target keyword: {exact query}
Reader emotional state: {frustrated / curious / comparison / ready}
Past headline that performed (tone anchor): {paste}
Existing top-ranking SERP titles to differentiate from: {paste 2-3}
Platform: {Google / Substack / LinkedIn / Twitter}
Banned words: {your AI tic list}
Frameworks to use, one per variant:
1) Question (answers a specific reader question)
2) Listicle with number (3 / 5 / 7 specifics)
3) How-to (with the outcome named in the title, not "how to do X")
4) Contrarian (reverses a default assumption the reader holds)
5) Status / identity (signals who this post is for)
For each variant return:
- The headline
- The reader outcome it promises (one sentence)
- Why it matches the reader's emotional state
- One framework-specific risk (e.g., "the contrarian only works if the reverse is genuinely defensible")
End with: which 1-2 to test first against the live SERP and why.Shorter variant — rewrite one weak headline
Below is the headline I drafted: {paste headline}.
Article takeaway: {sentence}. Target keyword: {keyword}.
The headline's specific weakness: {generic / no specificity / no reader benefit / keyword stuffed / boring}.
Return 3 rewrites that fix the specific weakness without changing the angle. Then 1 rewrite that changes the angle entirely.Sample output
A good 5-variant set for the takeaway “the right headline framework can 2x CTR on the same article”:
- Question: “Why does my blog get traffic but no clicks? The headline.” — pain-aware, matches a frustrated searcher.
- Listicle: “5 blog headline frameworks that doubled my CTR (with the data).” — number + proof + specificity.
- How-to (outcome-named): “Write a blog headline that beats the SERP — without keyword stuffing.” — promises the specific outcome.
- Contrarian: “Stop A/B testing your blog headlines. Do this instead.” — reverses a default; works only because the body genuinely argues against A/B-then-pick.
- Status / identity: “How serious indie writers test 5 headlines before publishing.” — signals the audience and frames the practice as expected.
A useful first-test recommendation: “Test the contrarian and the listicle first. The current SERP for ‘blog headline variants AI’ is dominated by 5 how-to variants, so the contrarian has white-space opportunity. The listicle works as the safety pick if the contrarian doesn’t land within 48 hours.”
How to refine
- If all variants feel similar: “Each variant must promise a different reader outcome, not just a different style. Rewrite any variant whose promise overlaps with another’s.”
- If keyword feels stuffed: “The keyword should read as the natural noun phrase a reader would say out loud, not as an SEO anchor. Rewrite so the headline survives the read-aloud test.”
- If contrarian feels forced: “The contrarian only works if my body genuinely argues against the assumption it reverses. If the article does not back the reversal, downgrade to a different framework.”
- If the variants are all too clever: “Reader benefit must be visible in 1 second. Cut clever framing that requires the reader to think. The headline competes on a phone screen against 9 other titles in the same SERP.”
- If the title and the H1 are identical: “Differentiate: the headline is the SEO/social hook (curiosity), the H1 is the in-article promise (clarity). Rewrite the H1 to be the version that delivers what the headline promised.”
Common mistakes
- Optimizing only for keyword density: Google rewards clarity and reader-intent match over raw keyword cramming, and its own title-link guidance warns that it will rewrite or ignore a title it judges keyword-stuffed or low-quality; a stuffed headline reads spammy and gets passed over.
- Writing the headline before the article: you do not know your best takeaway until the body is done; the inverse order produces forced articles that bend to the title.
- Picking by personal taste: your audience is not you; pick by reader emotional state + SERP white space, not by what makes you smile.
- Same framework as the top 3 SERP results: you are competing for clicks; matching format means matching commodity status. The opening is in the framework no one else used.
- Ignoring the SERP sanity check: five variants generated in a vacuum; the live SERP shows you which framework is overused vs underused, which changes the answer.
- Length creep: past roughly 580 pixels (about 55-60 characters, fewer if the title is heavy on wide letters) Google truncates with an ellipsis right where your benefit lives. Check the live preview in a SERP-width tool before publishing.
- Forgetting platform-specific previews: what fits Google may get truncated on LinkedIn or look unfinished as a Twitter card; if you publish to multiple platforms, test each preview.
- Using AI-tic intensifiers: “ultimate,” “unlock,” “elevate,” “supercharge” are the new “you won’t believe”; readers learned to skip them, and the algorithm penalizes them on some platforms.
FAQ
- Should the H1 match the headline exactly?: Close, but not identical. The headline is the SEO and social hook (the curiosity); the H1 is the in-article promise (the clarity). A headline of “Stop A/B testing your headlines” pairs well with an H1 of “A simpler way to pick a blog headline.”
- How long should headlines be?: Google truncates on pixel width, not character count, so the safe target is under about 580 pixels on desktop, which is roughly 55-60 characters of typical text (fewer if you stack wide letters like W and M). Keep social card titles under 90 characters. Shorter is not always better: “5” gets fewer clicks than “5 frameworks that doubled my CTR” on the same article.
- Should I run an A/B test on headlines?: Only if you have meaningful traffic. To detect a 10-20% CTR lift at 95% confidence on a typical blog, you need on the order of 1,000+ sessions per variant, and the test ties up a lower-CTR variant for one to two weeks. Below that volume you are reading noise. For most indie blogs, picking the best of 5 variants by a live SERP sanity check beats running underpowered A/B tests. Google does not offer a built-in title-tag split test, so headline A/B work usually means a tool like Google Optimize’s successors or a self-rolled meta-title rotation.
- What if the contrarian headline is the most clickable but the article doesn’t fully back it?: Pick a different framework. The bounce rate from a contrarian title that doesn’t deliver kills the ranking faster than a moderate framework with high satisfaction.
- How does this change for newsletter subject lines vs blog titles?: Subject lines optimize for open rate (curiosity, time-sensitive, sender trust); blog titles optimize for SERP click-through (intent match, clarity, differentiation). Different jobs, different prompts.