Profile authority rewrite: two readers, opposite appetites, one 2,600-character budget

---
name: profile-authority-rewrite
description: Rewrites a professional profile so it works on a human buyer and an automated topic classifier at the same time, which is a conflict rather than a synergy. Covers the published field caps and the separate truncation points, the roughly 75/25 divider split of the About section, the six-beat order of the human block, the repel pass, the identity-label dilution and label-count rules, a decision rule for choosing the primary label with a tie-break for when you cannot tell, and the single-sitting edit order. This skill should be used when a LinkedIn or similar professional profile is being rewritten, when a headline or About section is about to be replaced, when someone is changing field and the profile still describes the old one, or when a profile draws visits but no messages.
---

# Profile authority rewrite

## The claim this skill is built on

A professional profile is read twice, by two readers who want opposite prose, and almost every rewrite serves only one of them.

The first reader is a person deciding in about four seconds whether you are worth a message. That reader wants short sentences, one idea per line, white space between every thought, no compound clauses, and something specific enough to react to.

The second reader is automated. It is deciding what this account is about, so it can place you in the right retrieval sets, recommendations and adjacency rails. It wants the opposite: density, dated history, repeated domain nouns, named credential issuers, named sectors, named methods. It has no patience threshold. It has an evidence threshold.

**The obvious approach is to write for the human and hope the classifier copes. It fails invisibly.** The classifier runs first, and decides whether a person is ever shown your profile in a search result or a suggested-connection rail. A profile a human loves and a machine cannot categorise is read almost exclusively by people who already knew your name, which is the population that needed it least. From the inside it looks fine, because everyone who mentions it is complimentary. The reverse failure is easier to spot: the keyword slab that ranks and repels.

**The resolution is not compromise. It is segregation.** Both readers are served inside the same field, in different parts of it, in a fixed proportion, separated by a visible divider. Nobody scrolls past the divider, and that is the point of it.

One honest note first. LinkedIn documents none of its retrieval or ranking behaviour, so the mechanism above, that profile text is machine-read to infer a topic, is inferred from observed behaviour rather than published. Any claim about how that reading is weighted is a community hypothesis, worth checking against your own analytics before you trust it. The character caps below are a different matter, and are stated as facts.

## Part one. The two verdicts, written before anything is edited

Write one sentence twice, before touching a field.

**The current verdict.** In one sentence, what would a reader conclude from what is on the page right now? Write it only from the page, not from what you know about yourself or work you did and never listed. An invented example: "an engineer who started a marketing company."

**The target verdict.** What do you need a reader to conclude instead? "A practitioner working at the intersection of engineering and marketing with six years on both sides."

**Every edit either moves the current verdict towards the target verdict, or it is cut.** That is the acceptance test for the whole rewrite, and it is what stops the job becoming an afternoon of rewording, which is what it becomes by default, because rewording feels productive and moves the verdict not at all. Re-run the current verdict against the finished draft. If it has not moved, nothing you did mattered however much better the prose reads.

## Part two. The budget: caps, truncation points, and which one you write to

These are field limits on LinkedIn, verified as of August 2026. Check them yourself before budgeting, because fields change and a number in a file rots silently.

| Field | Hard cap |
|---|---|
| Headline | 220 characters |
| About section | 2,600 characters |
| Each experience description | 2,000 characters |
| Job title within an entry | 100 characters |
| Skills | 50 total, five of which can be pinned as top skills |
| Education description | 1,000 characters |

**The cap is not what you write to. The truncation point is.** Confusing the two is the most common budgeting error.

- **The About section collapses behind a see-more control after roughly the first three lines on desktop web, and fewer in the mobile app.** There is no documented constant and there cannot be, because the collapse depends on viewport width and font size. As a working budget, assume about 265 characters survive on desktop and about 180 to 200 in the mobile app, then write so the first 180 stand alone. Verify on your own phone, which takes fifteen seconds and beats any printed figure.
- **The headline is stored at 220 characters and almost never displayed in full.** In a feed byline, a search result row, a comment, a connection request and a notification it is clipped to one line at that surface's width. Treat the first 40 characters as the only portion guaranteed to appear everywhere and put your two most important domain nouns inside them. Characters 41 to 220 are seen by some surfaces and by every machine reader.
- **Experience descriptions collapse the same way.** First line carries the claim, the rest corroborates.

**Draft outside the field**, in a plain text file with a character counter, on whatever operating system you use, then paste. The counter inside the field only tells you that you are already over, at which point what gets cut is whatever sits nearest the cursor.

## Part three. The 75/25 split and the divider

The About section is the only field big enough to hold both readers, so it is the field that gets split.

**Roughly the top 75 percent is the human block**, anywhere from 70/30 to 80/20, which on a 2,600-character cap is about 1,800 to 2,000 characters of human prose.

**Below a visible divider, a line of four hyphens or a short plain label, goes the machine block.** About 500 to 700 characters of dense labelled lines rather than sentences: domain history with dates, previous role domains with dates, credential names and issuers, sector list, method list, the topics you write about. No connective prose, no attempt to read well.

**The rule that makes the split work: write the human block first to its own target length, fill the remaining characters with the machine block, and when the total runs over the cap, trim the machine block. Never the human block.**

The reason is degradation. The machine block degrades gracefully, because dropping the fourth sector costs a little evidence on one topic and nothing else. The human block does not degrade, it breaks: it is a six-beat structure, and removing a beat does not shorten it, it changes what it is. The usual casualty is the first beat, which is the shortest and looks like a throwaway line, and losing it turns a filter back into a brochure.

## Part four. The six beats of the human block, in order

The order is load-bearing. Each beat is separated by a blank line, and no beat is a compound sentence.

**1. Disqualify, in line one.** Name who this is not for, or name a state of affairs only your buyer is in. A qualifying first line doubles as the repel mechanism, and it has to sit inside the pre-collapse window, so budget it under about 120 characters. "You are not looking for a dashboard. You are looking for a price that holds through a bad quarter."

**2. Agitate, in their language.** Name the change that happened to them, not the thing you sell, using the nouns they use in their own meetings. Two or three short paragraphs. Here a reader either recognises their own situation described accurately by a stranger, the strongest single move this format allows, or does not.

**3. Bridge, in one line.** One sentence establishing you have sat in every relevant seat around the problem. Not a biography. A seat list.

**4. Prove, as a vertical stack, one metric per line.** The stack is skimmed rather than read, so the eye needs exactly one number per line to catch on, and a number inside a sentence is a number nobody finds. Five to seven lines. **Close the stack with the most disarming line available**, usually something you do not have: no certification, no funding, no team. A stack of only strengths reads as a claim. A stack with one admission at the bottom reads as an inventory.

**5. Offer.** One thing, described by what it ends with rather than what it consists of. One destination.

**6. Act.** One link. Offer and link at the end, never sprinkled through.

## Part five. The repel pass

Run this as a separate explicit step once the six beats exist, not as an attitude while writing them.

Write down the visitor you do not want: their stage, their budget, the thing they will ask for. Then write the exact phrases that will make that person close the tab. Four families that work:

- **An opening that names a stage they have not reached.** "By the time you have four planners and a pricing committee" excuses everyone smaller without insulting them.
- **A phrase only someone with your tenure would say.** Unremarkable to a peer, unfamiliar to a novice. It cannot be a claim about your tenure, it has to demonstrate it.
- **Vocabulary from one professional world rather than another.** The same job described in enterprise vocabulary and in startup vocabulary sorts two different populations, and both are true descriptions.
- **The deliberate absence of the words the wrong visitor searches for.** This one has a real cost, because those words are also evidence for the automated reader. Write down which topics you are giving up rather than discovering it later.

**A profile that repels nobody converts nobody.** The reply decision is comparative: your reader is looking at several profiles and picking one, so a profile built to be acceptable to everyone loses on the only ground that decides them, which is whether this person is obviously for me. It helps the second reader too, because acceptable-to-everyone produces a wide flat topic distribution and you are nobody's strongest match either. The repel pass serves both readers at once, which is why it survives when the budget gets tight.

## Part six. Identity labels, dilution, and why every extra label costs you

**The dilution rule.** A label can be perfectly accurate and unusable in the identity position, because an adjacent group has changed what it evokes. Your buyers may literally be agency owners and still refuse "agency owner" as an anchor, because a younger cohort made the term feel junior to them.

Test it: say the label aloud to three people in your target audience and ask what kind of person they picture. If the picture is younger, smaller or less serious than you are, the label is diluted in the anchor position. The fix is not a synonym, it is to substitute tenure and evidence markers, "you have been doing this for years", "you have real clients and a reputation to protect", and to rotate descriptors rather than lean on one. **A diluted label is still fine deeper in the body as a casual descriptor. It is poisonous only in the identity position.**

**The label-count rule is the less obvious half. Every additional identity label you claim weakens every other one**, for two separate reasons.

On the human side, a reader given three labels retains none. A stacked identity line reads as someone who has not decided, and each claim becomes less credible rather than more, because a person who is four things is assumed to be moderately good at four things.

On the machine side the arithmetic is starker. The evidence in your profile is finite: a fixed number of dated roles, credentials, skills and sector mentions. Each label you add divides that same evidence into more piles. Three claims backed by a third of the evidence each win none of the three topics, where one claim backed by all of it might have won one.

**So: one primary label in the anchor position, at most one qualifier attached to it, everything else expressed as evidence rather than as a label.** "Retail pricing and demand forecasting" is a label with a qualifier. "Consultant, speaker, investor, founder, coach" is five claims and no evidence for any of them.

## Part seven. The decision rule for the primary label

For the common case: someone genuinely does two things and has to anchor on one.

- **One domain holds more than about two thirds of your dated, verifiable evidence**, meaning years in role, credentials, published work and quantified results. That domain is the primary label, the other becomes the qualifier. Evidence decides, not preference.
- **The two are balanced, and buyers in one of them pay for the other.** Anchor on the market you sell into, use the second as the differentiator. The label names the room you want to be found in, the differentiator is why you win once you are in it.
- **Both are balanced and both are markets you sell into.** Anchor on the one where you can put a dated, quantified proof point in the top third of the About section. The label with proof under it is the one a human believes and the one the automated reader can corroborate, and an uncorroborated label is a weaker claim than no label.
- **The label you would pick fails the dilution test.** Keep the position and substitute tenure and evidence markers. Do not retreat to your second-choice domain, which trades a wording problem for a positioning problem.

**If you cannot tell, run the mirror test.** Take five fragments as the profile stands: the headline, the first 200 characters of the About section, and the titles plus descriptions of the three most recent roles. Paste them into a general-purpose model with no other context and ask one question: "In one sentence, what is this person's field, and what is your second guess?" Do it three times in three separate sessions so the runs are independent.

- **The three sentences disagree.** There is no primary label yet, and the fix is evidence rather than wording. Go to part eight and add the missing entry before writing another draft.
- **The three agree and they are wrong.** The label you have to displace is now named for you. Keep writing towards the target verdict and add dated roles and credentials on the target side until the corroboration outweighs the incumbent.
- **The three agree and they are right.** Stop rewriting. Ship it.

The mirror test reads a general model rather than the platform, so it is a proxy and should be described as one. It is still the only cheap, repeatable read of the second reader you can run yourself.

## Part eight. Work history as evidence for the claim, not as a job record

**Add the missing entry**, usually the highest-yield edit available. Almost everyone claiming a new domain has done unpaid, side or freelance work in it that never reached the profile: a volunteer role, a community project, a side practice, work for a trade body. Add it as a real entry, dated, in its correct chronological slot, quantified per channel. It converts a claim into a dated record, which is the one thing a rewrite cannot manufacture.

**Re-language the existing entries into the vocabulary of the claimed domain, and change nothing else.** The same enterprise pre-sales role described as go-to-market, positioning, enablement and messaging work is the same role, honestly described, in the vocabulary you are claiming. **Change the nouns. Never touch the numbers.** Inflating a metric during a re-language pass turns a rewriting job into a false statement on a public record former colleagues can read.

**Tag each entry with the domain skills** so they appear across several entries rather than once, which is the difference between a claim made in one place and a claim corroborated across a history.

**Then check credential balance.** If you claim an intersection of two domains, both have to be visibly credentialed. Count them on each side. Ten on one side and zero on the other while claiming both is an unsupported half, and the usual fix is not a new certificate, it is surfacing an existing one that was never listed.

## Part nine. The sitting: edit order, and why it is one sitting

**Order: additions, then rewrites, then the About section, then skills, then the headline.** New entries go first, because later fields refer to them and the About section should be written against the history that will exist rather than the one that does. Existing descriptions second. The About section third, once the evidence beneath it is settled. Skills fourth, tagged against the entries that now exist. The headline last, because it compresses everything above and compressing a moving target wastes the pass.

**Do all of it in one sitting.** Every save is a separate update event, so spreading a rewrite across a week produces a string of them where batching produces one. Editing after publication is widely reported to affect how content is treated on this platform, and the magnitude is unverified, so do not plan around a number. The reason that stands regardless of any ranking effect is plainer: a half-rewritten profile is publicly self-contradictory until you finish, with a new About section sitting above role descriptions that still describe the old field, and every visitor in that window reads the contradiction.

Budget 90 minutes to two hours, with the copy already drafted and counted outside the platform, and decide deliberately whether you want your network notified before you start rather than afterwards.

**Anything requiring an outside party is scheduled after the batch, never inside it.** Certificates that need issuing, endorsements and recommendations arrive on someone else's timetable, and putting them inside the sitting is how a two-hour job becomes a fortnight of half-finished profile.

## Worked example, compressed

An invented case. A freelance data consultant with eight years in retail analytics wants to be found for retail pricing and demand forecasting rather than for general dashboard work. Nobody here is real.

**Before.** Headline, 67 of 220 characters: "Freelance Data Consultant | Analytics | SQL | Python | Open to work". About section, 408 characters, one paragraph, opening "Experienced data consultant with a passion for turning data into insights" and closing "Currently open to new opportunities." Two of three roles carry no description at all. Nine skills, all tool names.

**Current verdict**, mirror test run three times: "a generalist data analyst who builds dashboards", "a business intelligence contractor", "a data analyst". Three runs, three different sentences, so by part seven there is no primary label yet and the first fix is evidence rather than wording.

**Target verdict:** "a retail pricing and demand forecasting specialist with eight years inside retail, six of them on pricing."

**Evidence first.** A two-year unpaid pricing project for a retailers' trade group, never listed, goes in as a dated entry in its correct slot. The grocery and fashion roles are re-languaged from reporting vocabulary into pricing and forecasting vocabulary, every number left exactly as it was. Skills drop from nine tool names to five pinned domain skills, tagged across all four entries.

**Headline after, 214 of 220 characters:** "Retail pricing and demand forecasting for mid-market retailers. Eight years in retail analytics, six of them in pricing. I build the forecast, then hand your team the model they run without me. Grocery and fashion." The first 40 characters, all some surfaces show, read "Retail pricing and demand forecasting fo", so both target nouns clear the clip.

**About after, 2,575 of 2,600 characters.** Human block 1,881, machine block 692, a 73/27 split. Line one is 98 characters and clears the collapse window on a phone: "You are not looking for a dashboard. You are looking for a price that holds through a bad quarter." The proof stack runs seven lines, one number each, closing on "No pricing certification. I learned this inside the business, not in a classroom." Below a four-hyphen divider sits the machine block: dated domain history from 2018, a seven-term domain list, a method list, five sectors, education and certificate names, two topics written about.

**The overrun, which is the part worth watching.** The first assembled draft came to 2,738 characters, 138 over. The trim came entirely out of the machine block, two sectors and one method line, taking it from 830 to 692. The human block was not touched.

**Verdict: ship it.** The mirror test now returns the same sentence on all three runs and it is the target sentence. One thing is unresolved and named rather than hidden: the profile claims pricing and forecasting and carries zero credentials on the pricing side, which the proof stack turns into an asset for a human reader and which does nothing at all for the automated one. A named certificate is the next action, scheduled after the sitting rather than inside it.

## Failure modes

**The keyword slab.** The whole About section is written for the machine, no divider, no beats. It appears in searches, collects profile views, receives no messages, and gets reported internally as a traffic success.

**The warm ghost.** The opposite. Beautifully written, entirely human, no dates, no sector nouns, no method names. Everyone who reads it is complimentary, and everyone who reads it already knew the person, because nothing gave an automated reader enough to place the account.

**The divider that reads.** The block below the divider is written in polite sentences instead of dense labelled lines, so it looks like the profile carried on and got worse. A human who scrolls judges the page by the weakest prose on it, and the machine block gains nothing from the connective words.

**Trimming the human block.** The draft runs 200 characters over, the writer cuts the shortest thing, which is beat one, and the profile stops filtering. What is left reads as a brochure opening on an agitation nobody has been qualified into.

**The five-label anchor.** "Consultant, speaker, investor, founder, coach." The reader retains none of them, and the same finite evidence now supports five claims, so nothing on the page is strongly associated with anything.

**The diluted anchor.** The label is accurate, checks out against the evidence, and evokes someone ten years more junior than the person using it. Buyers place them a tier down without articulating why, so the profile underperforms for a reason no reader can report.

**The drip edit.** The rewrite is spread across nine days. Several separate update events, and for nine days the About section describes one field while three role descriptions still describe another, in public.

**Metrics in prose.** The proof exists but lives inside sentences. A skimming eye finds no numbers, concludes there are none, and the strongest evidence on the page is never read.

**The unsupported intersection.** Two domains in the headline, ten credentials on one side and zero on the other. Attention goes to the credentialed half and the intersection claim collapses back into the domain you were already known for.

## What this skill does not do

- It does not know how any platform ranks, retrieves or classifies a profile. No weighting is published anywhere. The mechanism is inferred from behaviour, the character caps are checkable and stated as facts, and everything in between is a hypothesis to test against your own analytics.
- It cannot supply evidence you do not have. If the dated work, the credential and the number are absent, the honest output is a shorter and more specific profile, not a more confident one.
- It cannot tell you where the collapse falls on your reader's screen. The caps are fixed, the truncation point moves with viewport, font size and app version, so the first-line budgets here are working figures to verify on a phone rather than constants.
- It does not decide the positioning. If you cannot name the buyer and what they would otherwise do, this rewrites the wrong sentence very carefully, and it will read well while doing it.
- It does not handle anything requiring another person. Recommendations, endorsements and certificates sit outside the single sitting on purpose, because they run on other people's timetables.
- It does not measure the outcome. The two verdicts and the mirror test tell you whether the text changed what a reader concludes. Whether that produced messages is a question for your own profile analytics over the following quarter.