A stranger who has read only the title should be able to say what the piece is about, and the title should name something that exists and can be pointed at.
We have not measured this one. It is published because both of its tests have a pass or fail answer, which is unusual for anything about headlines.
What it knows, concretely. First, the object test, which is sharper than the familiar advice to be specific, because it comes with the list of words that feel like a concrete object and are not: a category, a decision, a process, a strategy, a mistake, a lesson, an insight, a mindset, a shift. A title whose head noun is any of those is naming an abstraction, and abstractions do not survive being seen alone in a feed, a search result or a bookmark list. Second, the three opening shapes that fail for one shared reason: metaphor-first, aphorism-first and wordplay-first all require the reader to already care in order to decode them, which is precisely backwards for a title, whose job is to create the caring. Third, the subtitle transplant test, which is that if the subtitle would sit comfortably under a different article, it is doing no work, together with the rule that the subtitle's object must differ from the title's object or the pair is one sentence split in half.
It also carries the two length limits and, more usefully, the fact that they are different problems. A search result title is truncated by pixel width, so front-loading matters more than counting characters. An on-page headline is not truncated at all and is instead competing with the first paragraph for the reader's next three seconds.
Who it is not for. If you are naming a book, a product or a campaign, these tests will make you literal, and those artefacts arrive with context the tests assume is missing. If your readers arrive by byline or by subscription, the title is doing less work than the tests assume. And a piece with a genuinely abstract subject can fail the object test honestly, which is why the file includes counterexamples rather than treating the tests as rules to obey.
Measuring one skill honestly costs about twenty model sessions: five runs with it, five without, on real material, each output graded alone by a session that is not told the other arm exists, against a rubric written by somebody who never saw the skill. We have not spent that on this one yet, so it ships labelled rather than ships silently.
How it would be measured. Tier A spine with a Tier B component. Both runs title the same ten drafts, five of which have a genuinely abstract subject and are therefore hard to title concretely. The objective half checks each output title mechanically: is its head noun on the non-object list, does it front-load the subject inside 60 characters, and does the subtitle's object differ from the title's. The subject test itself needs blind readers who see only the title and are asked what the piece is about, scored on agreement between readers.
Half of this is mechanically checkable and half is not. Checking whether the head noun of a title is on the non-object list, whether the subject appears within the first 60 characters, and whether the subtitle introduces a second object are all decidable from the string alone. That half could be scored without judgement.
The subject test is the problem. It requires readers who see only the title and report what they think the piece is about, and agreement between them is the actual measurement. That means recruiting people, keeping the drafts away from them, and accepting that a small panel gives noisy answers on a task where the interesting cases are the ambiguous ones.
The material also has to include hard cases. Ten drafts with obviously concrete subjects would let any approach pass, so half the set needs a genuinely abstract subject where the honest answer may be that no concrete title exists.
The rule that decides pass or fail was written down before any run was executed and it does not move afterwards. It is in the method note on the hub, along with the full results table including every skill that was tested and cut.
Stated plainly, because a skill that claims everything is useful for nothing.
name and description in the file's frontmatter, so you can also invoke it by name.An editor who writes headlines for a living is better than this, and the gap is not close. Newsroom headline writing is a specialism with decades of feedback behind it.
Headline scoring tools are cheap and fast and answer a different question. They compare your string against patterns that have performed, which is genuinely useful and easy to over-fit: a title optimised to score well can still leave a stranger unable to say what the piece is about, because the scoring never asks.
For the search half specifically, the platform documentation beats any summary of it, including this one, and it changes more often than a file can track.
The model with no skill at all will produce competent, plausible headlines on request, and for most posts that is enough. What this file adds is a pair of refusal tests, one on the subject and one on the object, that a title has to pass before it is offered at all. Their whole function is to make a beautiful title that names nothing fail against a stated criterion rather than against somebody's taste.
--- name: headline-and-title-test description: Tests a title, headline or subtitle against two falsifiable checks: whether a stranger who has read only the title can say what the piece is about, and whether the title names something that exists and can be pointed at rather than an abstraction. Includes the explicit list of words that feel like objects and are not, the three opening shapes that fail, the subtitle transplant test, specificity mechanics that actually move a title, and the different length limits for a search result and an on-page headline. This skill should be used whenever a title, headline, subject line or subtitle is being written or argued about. --- # Headline and title tests ## The claim this skill is built on Most arguments about a headline are arguments about taste, and taste arguments are settled by whoever is more senior or more tired. Two tests turn the same argument into something with an answer. The obvious approach, which is to write several candidates and pick the one that feels strongest, fails for a specific reason. You cannot feel a title the way a stranger does, because you have read the piece. Every title looks clear to the person who knows what it refers to. The tests exist to remove that advantage. Both tests assume the hardest condition, which is also the normal one: the title travels alone. It appears in a search result with no context, in a feed between two unrelated things, in a bookmark list six weeks later, in a chat message with no preview. It does not appear next to your introduction, which is the only place it looks clear. ## Test one, the subject test **A stranger who has read only the title can say what the piece is about.** Run it: show the title to two people who have not read the piece. Ask one question, "what do you think this is about?", and write down both answers. - Both answers match each other and match the piece: pass. - The answers differ from each other: fail. The title is ambiguous, and ambiguity in a title is not intrigue, it is noise. - The answers match each other but not the piece: worse than a fail. The title is legible and wrong, which means you will get readers who bounce and readers who felt misled. The usual cause of failure is that the title names a feeling rather than a topic. "The Quiet Cost of Moving Fast" produces different answers from different readers because it names an attitude towards a subject instead of a subject. Feelings do not disambiguate. ## Test two, the object test **The title must name something that exists and can be pointed at.** This is the sharper of the two, because it can be run alone, in a second, without recruiting anybody. Take the head noun of the title and ask whether you could photograph it, open it, run it, sign it, or put it on a table. An object is a thing: a spreadsheet, a signup form, a database migration, a rota, a refund policy document, a seven-screen checkout, a Tuesday deploy, a support queue, a pricing page, a 40 page contract. **The list of things that are not objects, even though they feel like one.** A title built on any of these is naming an abstraction: - a category - a decision - a process - a strategy - a mistake - a lesson - an insight - a mindset - a shift Each of these is a container for an object rather than the object. "The Decision That Changed Our Roadmap" names a decision, which is a container. What is inside it, the thing you could point at, is a roadmap with one item removed, or a contract signed, or a team of four moved onto something else. Put that in the title. The same applies to the verbs that pair with them. "Rethinking", "reimagining", "unpacking" and "the case for" all describe an activity performed on an abstraction and add no object of their own. ## The three banned opening shapes Three shapes fail for one shared reason: each requires the reader to already care in order to decode it, and caring is what the title is supposed to produce. **Metaphor-first.** "Sailing Into the Fog". The reader has to work out what the fog is before they know whether the piece is for them, and they will not do that work for a stranger. A metaphor in a title is a bill you send before the goods arrive. **Aphorism-first.** "Slow Is Smooth, and Smooth Is Fast". It reads as wisdom, so it feels like a strong line, and it names nothing. Two different readers will attach it to two different subjects, so it also fails the subject test. Aphorisms work in the body, after the object has been established. **Wordplay-first.** "Table Manners: Getting Your Data to Behave". The pun requires recognising both meanings, which requires already knowing the subject. Wordplay is a reward for the reader who has arrived, not a reason to arrive. It also travels badly: it is the first thing lost in translation and the first thing that reads as noise when quoted flat. The pattern in all three is that they are enjoyable to write. That is worth knowing, because the pleasure of writing them is the reason they survive review. ## The subtitle Two rules, both testable. **The transplant test.** Take the subtitle and put it under three other articles on the same site. If it would make sense under any of them, it is doing no work and is describing a genre rather than a piece. "What we learned and how you can apply it" transplants under everything ever written. **The second-object rule.** The subtitle's object must be different from the title's object. If the title names the checkout flow and the subtitle also names the checkout flow, you have written one sentence and split it in half, and you have wasted the second half. The subtitle should add the thing the title left out: the constraint, the outcome, the number, the counterintuitive part, or the reader it is for. Worked pair: title "The seven-screen signup we cut to two", subtitle "What we lost by removing the address step, and who complained". Two different objects, and the second one raises the question the first one does not. ## Shape variety Check this title's shape against your last five published titles. Shapes include the how-to, the number list, the question, the against-X, the object-plus-outcome, the first-person account, and the flat declarative. If three of the last five share a shape, the shape has become a formula, and a recognisable formula stops carrying information: readers pattern-match it to a genre and skim past. This is a check on the sequence, not on the individual title, which is why nobody notices it. It costs thirty seconds and it is the only one of these checks that requires looking outside the piece. ## Specificity mechanics Four things reliably move a title from abstract to concrete. Each replaces an adjective with a fact. - **A number**, but a real one. "Seven screens", "40 pages", "four months". A number invented for the title is the fastest way to lose a reader who counts. - **A named constraint.** "Without adding a database", "on a two-person team", "with no downtime". Constraints are interesting because they imply the interesting part of the story. - **A timeframe.** "In one afternoon", "after three years", "the week before launch". Timeframes create scale, which is the thing readers use to judge relevance. - **A concrete noun in the head position.** Covered above, and the most reliable of the four. **On curiosity gaps, honestly.** A title that withholds the answer does raise click-through. It also spends trust, and the spend is only worth it if the piece pays out immediately. The rule: a curiosity gap is acceptable when the first paragraph closes it. If the reader has to reach the sixth paragraph to find what the title implied, they have been charged for something they already paid for, and the cost lands on the next title you publish rather than this one. ## Length, which is two different problems **In a search result.** The displayed title is truncated by pixel width, not character count, which is why character advice is only ever an approximation. Around 580 to 600 pixels on a desktop result is the working figure, which lands near 50 to 60 characters for typical mixed-case text and considerably fewer for a title full of wide capitals. Two consequences: front-load the object, because the front is the part guaranteed to survive; and treat the site name suffix, if your template appends one, as part of the budget. Also note that the title shown in results is generated by the search engine and may not be your tag at all, a behaviour that became common after the title generation change in August 2021. Verify with a live result rather than a character counter, as of August 2026. **On the page.** No truncation, so length is not the constraint. The constraint is that the headline is competing with the first paragraph for the reader's next three seconds, and a headline long enough to wrap onto three lines on a phone reads as a paragraph and gets skimmed like one. Two lines on a narrow viewport is the practical ceiling. **Related, and different again.** A social card title and an email subject line each truncate at their own points, and mobile mail clients commonly show only the first 30 to 40 characters. Do not reuse one string across all four surfaces and assume it works. ## The decision rule - Two blind readers give the same subject, and it is the piece's subject: **ship it.** - The two readers disagree: **fail.** Rewrite by naming the object, not by adding words. - The two readers agree on a subject the piece does not deliver: **fail, and rewrite from the piece.** This one is worse than ambiguity because it is a promise. - You cannot get two readers: **run the object test alone.** If the head noun is on the non-object list, treat the title as failed regardless of how well it reads. - You still cannot tell, and two candidates seem equal: **choose the more specific one.** A specific title fails visibly, by attracting the wrong reader who says so. A vague title fails invisibly, by attracting nobody, and you never learn it failed. ## Worked example, six titles The titles below, and the figures inside them, are invented for this example. **1. "The Hidden Cost of Technical Debt"** Subject test: readers say "technical debt", which is a category, not this piece. Object test: the head noun is "cost", an abstraction. Fails both. Rewrite: "We spent four months migrating off a database we picked in a weekend". Object: a database. Number: four months. Constraint implied. **2. "Rethinking Onboarding"** Object test: "rethinking" is a process, "onboarding" is a category. Two non-objects in three words. Rewrite: "The seven-screen signup we cut to two". **3. "Lessons From Our First Year of Support"** Object test: "lessons" is on the list. Subject test: readers agree it is about support, which is a category, so it passes weakly and identifies nothing. Rewrite: "The three questions that were 60 per cent of our support queue". **4. "Small Hinges Swing Big Doors"** Banned shape, aphorism-first. Names nothing. Two readers will give two subjects. Rewrite: "Adding one confirmation screen cut refund requests by a third". **5. "Table Manners: Getting Your Data to Behave"** Banned shape, wordplay-first, and the pun is only legible to someone who already knows the subject. Rewrite: "Why our export produced 12,000 rows and the report showed 11,984". **6. "A Shift in How We Think About Pricing"** Object test: "shift" is on the list, and "how we think" is a mindset, also on the list. Rewrite: "We moved from per-seat to per-project pricing and lost our two largest accounts". **Verdict:** all six failed the object test on the head noun alone, which took under a minute total and required no readers. Four of them also fail the subject test. The rewrites are longer in every case, which is the usual outcome and is not a cost, because none of them exceeds the pixel budget once the object is front-loaded. ## Counterexamples, so the tests are applied rather than obeyed The tests assume the title travels alone. Where context reliably travels with it, they relax, and pretending otherwise produces flat, literal titles that lose to better ones. - **A byline that carries an audience.** If readers arrive because of who wrote it, the title is not doing the recruiting and a metaphor can work. This is a privilege that has to be earned before it can be spent, and it is the most commonly assumed and least commonly held. - **A book.** The cover, the subtitle, the spine, the shelf, and the recommendation all arrive with it. Book titles routinely fail the object test and work. - **A subscriber newsletter.** The reader opened it because of the sender. The subject line's job is closer to signalling which week this is than to recruiting a stranger. - **A title where the object is a proper noun everyone already knows.** One word can be enough, because the word carries the whole subject. - **A piece whose genuine subject is abstract.** A philosophical essay about attention is about attention. Forcing a fake object onto it is worse than an honest abstraction, and the honest move is to accept the failure and win on the subtitle instead. ## Failure modes **Judging your own title from inside the piece.** Everything reads as clear when you know the referent. This is the failure the subject test exists to remove, and the reason the substitute self-check is weaker. **Adding words instead of naming the object.** A failed title gets longer rather than more concrete, and now it fails at 90 characters. **Keeping the working title.** The file name becomes the title because nobody scheduled the decision. Most placeholder titles ship not because someone chose them but because nobody chose anything. **Mistaking ambiguity for intrigue.** Two readers giving two different answers is not a hook, it is a defect that reads as a hook to the author. **Repeating the title in the subtitle.** One sentence split across two lines, with the second line adding no object and therefore no reason to keep reading. **Reusing one string on four surfaces.** The search result, the on-page headline, the social card and the email subject truncate at different points and have different jobs. **A curiosity gap the piece never closes.** The click arrives, the trust leaves, and the damage lands on the next thing you publish. **Formula drift.** Every title this quarter has the same shape, each one defensible alone, and the sequence has become a genre the reader skims. ## What this skill does not do - It does not predict clicks. It tests legibility, and legibility and click-through are related but not the same property, so a title can pass both tests and underperform a vaguer one. - It does not do keyword research. Whether the words in your title match search demand needs a keyword tool and real volume data, and this file has neither. - It does not write for social platforms specifically. Each has its own truncation, preview behaviour and norms, and those change faster than a file can track. - It cannot run the subject test on its own. Without two blind readers you get the object test and a weaker self-check, and the file will say so rather than pretending the substitute is equivalent. - It does not apply cleanly to names. Products, companies, campaigns and books carry context these tests assume is absent, and applying them there produces descriptions rather than names.
These skills all ask your assistant to check things against your actual codebase, your actual schema, your actual design system. Locul keeps that context current on its own, from the files you already have, on your machine. Mac and Windows, free to start.