Comet is Perplexity's AI browser, and its built-in assistant can read your LinkedIn profile the way a stranger does, then grade it. AI Central's walkthrough pairs the browser with one long prompt that scores six profile elements out of sixty, applies automatic fail conditions, and returns a named fix for each section. The setup needs only a LinkedIn profile and a free download. The value is not the score, it is how specific the rubric is about failure.
Reviewed 7 August 2026.
Why a browser agent is the right tool for this job
Most LinkedIn profile advice is generic because whoever wrote it has never seen your profile. Comet closes that gap by putting the model inside the browser session where your profile is already loaded and already logged in. AI Central's guide, Fix Your LinkedIn Profile Using Comet, is blunt about why the browser is the interesting part.
Its key differentiator is a native AI agent, which can be utilized to optimize platforms like LinkedIn
That is AI Central's framing, positioning Comet as a competitor to Google Chrome rather than an extension bolted onto one. The distinction carries the technique. A chat window in another tab cannot see a page that sits behind a login. An assistant living in the browser can.
The setup, four moves
The mechanical part is four steps, and none of them are hard.
- Go to Perplexity.ai, sign in, then install Comet from the bottom-left corner. AI Central notes the install is free.
- Open Comet and either set up a profile or use a guest profile. You can import your existing Chrome setup.
- Log in to LinkedIn inside Comet and open your own profile page.
- Click Assistant in the upper right corner to open the side window, then paste the prompt.
That is the entire installation. Everything worth studying lives in the prompt.
The rubric is the product
The prompt opens by naming the job precisely instead of asking for feedback.
You analyze what happens in the first 3 seconds someone sees a LinkedIn profile and what makes them stay for 30 seconds.
That line, from AI Central's prompt, does most of the work. It replaces the useless ask, how is my profile, with two timed questions, and every scoring band underneath exists to answer them.
Six elements get scored, each worth up to ten points, for sixty total: photo, headline, featured section, about section, experience, and connection display. The six then split into two halves of thirty. Photo, headline and connections are the instant-click factors. Featured, about and experience are the stay-and-read factors. A profile fails the three-second test if it scores under fifteen on the first three, and the thirty-second test if it scores under fifteen on the last three.
It is built to fail you
Most audit prompts are flattery machines. This one begins with three conditions that end the evaluation before it starts.
No profile photo = 0/60 total score (stop evaluation)
AI Central applies the same treatment to a missing headline and to any account under fifty connections. Zero out of sixty, evaluation over.
The photo score is binary too, ten or nothing. Four conditions all have to be true: a human face filling at least sixty percent of the circle, a solid colour background, eyes on the camera, professional lighting. One failure zeroes the section. The about section is just as unforgiving at the bottom end, scoring zero outright if it opens with the words I am.
The design principle is worth taking even if you never open Comet. Rubrics that allow partial credit everywhere produce scores that drift upward and mean nothing. This one forces a verdict.
What a real audit returns
AI Central shows two runs, one strong and one weak, and the gap between them is where the technique proves itself.
The strong profile took twenty-four out of thirty on the instant-click half: nine for the photo, ten for connection count, five for the headline. The verdict was narrow and actionable.
strong profile but missing a curiosity gap in the headline.
The headline under review, quoted in AI Central's output, was Master AI before it masters you. It reads well. It still lost half its points, because reading well and opening a curiosity gap are different jobs.
The weak run is the more useful one. A profile with a photo, a job title in German and 149 connections came back at fifteen out of sixty. No featured section, zero. No visible about section, zero. Experience showed a title and a brief description with no numbers, five. Ten out of thirty on instant click, five out of thirty on stay-and-read, both timed tests failed. Nothing there is broken in a way anyone would notice. It is simply invisible, and the rubric says so in numbers instead of adjectives.
The finding that justifies the whole exercise
Buried in the output is a claim about LinkedIn's behaviour that changes how you should treat the headline field. When someone hovers over your name in a comment thread, AI Central reports that only your first thirty-eight headline characters and your photo appear, not your banner.
Meaning if those 38 characters don’t open a curiosity loop, your entire posting visibility is wasted.
That is AI Central's conclusion, and if it holds, the headline stops being a description of your job. It becomes the advertisement that runs every time you comment on somebody else's post. Thirty-eight characters is roughly six or seven words. Write them as though nothing else on the profile exists, because in that hover card, nothing else does.
Where to push back
Two things in the output deserve scepticism.
First, the assistant does not reliably obey its own rubric. The photo section is written as zero or ten with nothing in between, yet the sample output awards nine out of ten. In the same run the headline is graded against tests the prompt never specified: action verbs, a number in the first forty characters, an I statement. Defensible tests, improvised on the spot. If you want scores that stay comparable across runs, tell the assistant to hold to the stated bands.
Second, the fixes section demands a predicted percentage on every recommendation: expected click increase, expected view increase, expected engagement increase, expected profile save rate. No model can know those numbers for your profile. Read them as a ranking of which fix to do first, never as forecasts. The qualitative half of each fix is the half with value. When the weak audit prescribes a solid sky blue photo background, given as hex code 0099ff, that is something you can act on this afternoon.
One practical note before you paste. The connection display band defines a five-point rule and a zero-point rule but never states what earns ten, so the top of that section is left to the model's judgement. If you want a perfect sixty to be reachable, write the missing rule in yourself.
What to do with this
Four moves, roughly in order of return.
- Fix the headline first. It is one of the three instant-click factors, and it has to state a clear outcome inside forty characters to score full marks.
- Add a featured section if you have none. One to three tiles with a clear call to action takes you from zero to ten in a single sitting.
- Rewrite the first line of your about section to open with a number or a proof point, and close the section with a call to action.
- Point the same prompt at a profile you admire. AI Central suggests analysing leading profiles to identify the winning elements, and the rubric works unchanged on someone else's page.
The larger lesson is that this is not really a LinkedIn technique. It is a template for auditing anything you can put in front of a browser agent: a scoring system with hard fail conditions, a logged-in assistant pointed at a live page, and a demand for exactly one fix per section rather than a list of observations.
Do I need to pay for Comet to run this?
AI Central's walkthrough installs Comet at no cost, from the bottom-left corner of Perplexity's site after you sign in. A LinkedIn profile is the only other requirement listed.
Can I do this in Chrome instead?
The method is written for Comet because it depends on the assistant reading the logged-in page from a side window in the same browser. Comet does let you import your existing Chrome setup, so moving over does not mean starting from scratch.
What score should I be aiming for?
Sixty is the ceiling, split into two halves of thirty. Under fifteen on the first three elements fails the three-second test, under fifteen on the last three fails the thirty-second test. The strong example reached twenty-four out of thirty on its instant-click half.
Will it actually change my profile for me?
The walkthrough shows the assistant returning the analysis and the specific edits in the side window, section by section. Making the changes is still your job.