Headline optimization: grounding a guess in real data

A headline is a bet on which words people actually search for. Here is how AdAstra generates alternatives and then checks that bet against real keyword data, rather than just asserting one headline is better than another.

~6 min read . Headlines

A headline does two jobs at once: it has to describe what the piece is about, and it has to contain words a real person would actually type into a search box. Get the second part wrong and it does not matter how well written the headline is, almost nobody finds it. AdAstra treats every headline alternative as a testable bet on keywords, and grounds each bet in real Google Ads data rather than asking an AI model to simply assert one version is better than another.

Why headlines are keyword bets

Every headline implies a set of keywords, whether you chose them deliberately or not. "Five Ways to Save on Home Insurance" and "How Home Insurance Premiums Actually Work" cover similar ground but bet on different search phrases, with different volume and different commercial intent behind them. Optimizing a headline, in AdAstra's sense, means generating alternative phrasings and then checking which keyword bet each one actually places, using the same live Google Ads keyword data that grounds every other part of AdAstra.

This is why AdAstra's headline flow does not stop at generating clever text. A headline that reads well but implies keywords nobody searches for has not been optimized for anything measurable, it has just been rewritten.

Grounding a headline in real search data

AdAstra's headline optimization flow works one alternative at a time. For each alternative, an AI model first generates the headline text itself, tuned to your topic, your chosen optimization goal and how far you have allowed it to deviate from the original. It then identifies the main keyword or keywords implied by the headline it just wrote, and calls the same Google Ads keyword tool used elsewhere in AdAstra to look up real data for those specific keywords, for your target country and language.

That order matters. The headline is not scored against the keyword data you started with, it is scored against the keywords the headline itself actually contains. Two alternatives generated from the same topic can and do get graded on different keyword lookups, because a headline that shifts the angle also shifts which words carry its search intent.

Projected volume, CPC and percent change

Once the keyword lookup returns, each headline alternative reports a small set of fields together: a projected search volume (the absolute monthly search figure for the headline's own keywords), a projected CPC, and a percent change for both volume and CPC relative to a baseline. It also reports an overall 0-to-100 score for how well the headline serves the optimization goal you selected.

The volume and CPC figures come directly from the same live Google Ads lookup used throughout AdAstra: no separate estimate, no fabricated substitute. The percent-change figures and the score, by contrast, are the model's own assessment, informed by that real data rather than measured independently by the Ads API itself. It is worth keeping that distinction in mind: the absolute numbers are grounded in a live API call, the deltas and the score are a model's read of what those numbers mean for your goal.

That score is illustrative only, a reminder of the 0-to-100 scale, not a claim about any real headline. The same caution applies to any percent change you see next to a headline alternative: treat it as a directional read grounded in real keyword data, not as a guaranteed outcome.

The headline persona

AdAstra writes headlines the way a viral content strategist would. The model is briefed to maximize clicks and engagement, producing headlines that are enticing and curiosity-driven, optimized for search visibility and volume. That persona is the whole point of the flow: it is not just rewriting your title, it is reaching for the version most likely to win the click on a results page.

The CPC figure attached to each alternative is the publisher-side number, the revenue-share-adjusted figure covered in the keyword economics article, so the projection reflects what that click is actually worth to your site rather than what an advertiser would pay for it. The optimization goal is not just a label: it changes what the model is trying to accomplish and which figure it grounds the answer in.

Clickbait guardrails

Because these headlines are explicitly optimized to maximize clicks, there is a real risk of drifting into sensationalism or headlines that overpromise what the content actually delivers. AdAstra exposes a dedicated prevent clickbait toggle for exactly this reason. When it is enabled, the model is instructed to avoid sensationalism, exaggeration and emotionally manipulative language, and to keep the headline informative, accurate and a fair reflection of the underlying content.

This toggle is independent of the optimization goal and the optimization level (how far the alternative is allowed to deviate from your original topic). You can ask for an aggressive, high-deviation rewrite aimed at maximizing search volume and still keep the anti-clickbait guardrail on: the two settings answer different questions, how far to push the rewrite versus how the rewrite is allowed to sound.

The short version

• A headline implies a keyword bet whether you chose it deliberately or not; optimizing it means testing that bet, not just rewriting the text.
• For each alternative, AdAstra generates the headline first, then looks up real Google Ads data for that headline's own implied keywords.
• Projected search volume and CPC come from a live Google Ads lookup; the percent-change figures and the 0-to-100 score are the model's own assessment informed by that data.
• Headlines are optimized for clicks and volume, and the projected CPC is shown as the publisher-side figure: what the click is worth to your site.
• The optimization level controls how far a headline may deviate from your original topic, separately from the optimization goal.
• A real preventClickbait toggle exists to keep headlines accurate and non-manipulative, independent of how aggressive the rewrite is.