What your LinkedIn searches don't tell you
Across 143,258 profiles, why keyword search keeps surfacing the same candidates.
Why do your searches keep bringing back the same candidates?
In brief
- Across 143,258 public LinkedIn profiles analysed in 2026, almost all in France, one profile in two has written nothing in its About section, and 58% do not describe the role they currently hold.
- 20% of profiles describe none of their experiences. These are established professionals, nine years of experience at the median, who are barely active on the platform.
- On the keyword “python”, 53% of the profiles found are found only through a role they have already left, four years ago at the median.
- Job title, employer and dates are filled in nearly 100% of the time: searching by career path reaches the profiles that text search ignores.
A data clean-up that went sideways
A few months ago I set out to clean my database of LinkedIn profiles: bring it up to standard, classify it by occupation, deduplicate it. A long, tedious job I had been putting off for a very long time.
But from the very first counts, it veered off course. Auditing the database, I noticed that one profile in two had written nothing in its About section, and that nearly six in ten had not described the role they currently hold. Wary of my own data quality, I checked a sample of 300 profiles to find out whether this was a flaw in my collection or whether it came from LinkedIn itself. A few hours later, the verdict came in. What I had taken for a bug was not one: LinkedIn data is incomplete.
That first result led me to dig further in order to understand: 143,258 profiles, 952,242 experiences, boxes filled in and boxes left empty. In this article I share what I found, and why it changes everything for anyone hunting for their gem by typing keywords into LinkedIn’s search engine. The numbers are French. The mechanism has no reason to be, and the end of the piece gives you a way to check it on your own data.
A self-reported directory, not a register
A LinkedIn profile requires only three pieces of information per experience: a job title, an employer, a start date. That is the skeleton of a profile, and it is mandatory.
The rest? The About section, the description of each role, locations and contract type are all optional.
| Field | Filled in |
|---|---|
| Job title and company name | 100% of experiences |
| Start year | 99% |
| Company linked to its LinkedIn page | 85% |
| Role location | 80% |
| Experience description | 66% |
| Current experience description | 42% |
| Contract type | 49% |
| About section | 1 profile in 2 |
If LinkedIn knows where people have worked, it knows far less about what they did there. The description of the latest role is filled in only 42% of the time, and that figure shrinks to 34% if we keep only the profiles whose description runs longer than 150 characters. Which yields this kind of write-up: “Lead Product Manager at ‘My Company’: defining the product vision, driving the roadmap and optimising the user journey.”
Yet a keyword search relies on exactly that content: a job title rarely contains the name of a tool or a method; those words live in the descriptions, the About section and the skills, fields that nobody is required to fill in.
The structure of the data, its presence or its absence, decides who can be found before any query is typed.
The current role, the profile’s blind spot
The role that interests the recruiter is the one candidates describe the least: on LinkedIn, the current role is described in only 42% of profiles, against 72% for older roles.
| Experience | Described |
|---|---|
| Current role | 42% |
| Previous role | 58% |
| Older roles | 72% |
The more recent the role, the less it is documented: it is described only 39% of the time for people who joined their employer in 2025-2026, and 51% for those who joined between 2010 and 2017.
Among those who have written nothing about their current role, two in three have described an older one. So they know how to write, but chose not to describe the role they hold today. On starting a new job, the candidate adds a line with the title and the company, and does not come back to it for a while. Describing your current role means dressing the shop window, but staying silent is a reasonable choice for someone who is not looking for anything. The description of the current role will most likely be written the day its holder decides to make it known. If I dared, I would say the day they start looking elsewhere.
One profile in five says nothing
20% of the LinkedIn profiles studied describe none of their experiences: roles, employers, dates, and nothing else. Among those 20%, three in four have no About section either: that is 15% of the database, nearly one profile in six, without a single written line.
And contrary to what one might expect, these are neither juniors nor the very senior.
| Profiles with descriptions | Profiles with no text | |
|---|---|---|
| Years of experience (median) | 10 years | 9 years |
| Roles listed (mean) | 7.5 | 3.8 |
| Show only their current role | 0.6% | 25% |
| Master’s-level degree listed | 76% | 48% |
| Followers (median) | 637 | 217 |
Same seniority, half as many roles listed, a degree listed half as often, a network three times smaller: among profiles with fewer than a hundred followers, 58% have described no experience at all; above ten thousand followers, 6%. Unsurprisingly, these empty profiles belong to people who are barely active on LinkedIn: they are registered there, but they do not live there.
So these are established professionals, in post, rarely approached because nobody sees them: for direct approach, it is hard to imagine a better scenario.
Silence has a job title too. By occupational family, the share of profiles with no description at all varies threefold:
| Family | No description at all | Current role described |
|---|---|---|
| Product | 9% | 53% |
| Marketing | 9% | 53% |
| Consulting | 14% | 45% |
| Software | 16% | 45% |
| Data | 18% | 42% |
| Infrastructure | 20% | 40% |
| Finance | 22% | 39% |
| Engineering and manufacturing | 27% | 40% |
Those whose job is to make themselves visible write. Where the hunt is hardest, in industry, finance and infrastructure, the profiles are the barest.
Why we keep finding the same people
A LinkedIn keyword search surfaces first the profiles that have written something, and often for a role they have already left.
In the age of data, take a keyword everyone types: “python”. 24,699 profiles in the database contain the word “Python” somewhere: in the About section, the headline or the description of a role.
That is what a keyword search surfaces when it reads the text.
The question is where the word sits.
| Where “python” appears | Profiles | Share |
|---|---|---|
| In the current role, title or description | 5,313 | 22% |
| In the About section or headline only | 6,188 | 25% |
| In a former role only | 13,198 | 53% |
More than one profile in two is therefore found only thanks to a role its owner has left, four years ago at the median. On “java”, seven in ten, and the role dates back eight years; on “sap”, two in three, and it dates back seven years. The recruiter thinks they are reading a current skill. They are reading a CV that has aged.
And the 28,249 profiles that describe none of their experiences? For them, the word can only come from the About section, the headline or a job title. On “python”, that surfaces 682 of them: two in a hundred, against one in five among those who have described at least one role. LinkedIn also reads the skills people tick, and a “Python” box brings up a few more; but a ticked box says neither where nor when.
That is the mechanism.
Who fills in their profile?
The person who fills in their LinkedIn profile is first and foremost the person who wants to be seen, whether they are looking for a job, a client or a network. They took the time to fill in optional fields in order to exist on this network, to give themselves a chance of being discovered. The database confirms it: among the 3,095 profiles whose headline reads freelance, independent, available or open to work, 58% describe their current role against 42% for the others, and the silent ones are half as numerous.
When I source by keyword, I get these profiles back, again and again, and I end up approaching that same visible minority every time.
The others stay in the shadows. They are there, but nobody reads them.
The visible minority, for its part, gets everything else: the same messages, from the same headhunters, for the same roles, until the day it stops replying. The silent ones, whom nobody approaches, reply better. I have no proof of this beyond my own assignments, but it has never let me down.
So handing the job description to an AI to get better keywords moves nothing. The query gains synonyms, and that is useful: “chef de projet” does not see “project manager”, a language model knows that. But it still queries the same text, which is missing in the same place. AI rereads what is written. It does not bring into existence what is missing.
Reading the negative space
The candidates a keyword search cannot see are nonetheless on LinkedIn, on profiles where the text is missing but the structure holds. That structure is what we must read, then complete.
Start from career paths, not from keywords. Job title, employer and dates are there in nearly 100% of cases, among the silent as much as among the talkative.
“Who has worked at this company, at this level, since when” is a question the database can put to everyone.
In my training sessions, I keep repeating it to trainees: map the market before looking for the people. Teams, flows between competitors, bridges from one employer to the next can all be reconstructed without a single candidate having written a line.
Inferring what the profile does not say
A data engineer in a GCP shop knows GCP, whether they wrote it down or not. The environment we work in shapes our skills. The implicit can be read in the environment; it is up to us to uncover it, to bring it to the surface when it is not spelled out.
This problem is not specific to LinkedIn. A CV database, an ATS, my own database: any engine that searches text inherits the same void, and sometimes makes it worse.
Asking the tool to count what it cannot see is, by its very nature, impossible. In an ideal world, a search that displays “200 candidates” would say how many profiles within the same scope remain out of the text’s reach. The number exists in the database; no engine shows it.
What is left on the table
I have not finished classifying the database. But I now know who I am reading when I type a word: those who had a reason to write, at the moment they had it. The others are waiting to be read differently. Twenty-eight thousand profiles without a single line, and I do not yet know which one to start with.
Try it on your own data
The three counts that matter take about an hour on any export you already have: your sourcing database, your ATS, your CRM.
- The share of profiles with an empty About section, and the share with no description of the current role.
- The share of profiles with no description on any role at all. These are the people your text search will never surface.
- For the keyword you type most often, where the match sits: the current role, the About section or headline, or a role already left. The date of that role is the age of the skill you think you are reading.
If your figures differ from mine, I would like to hear about it. If they match, you know how much of your market your keywords leave untouched.
Where these numbers come from
Database: 143,258 public LinkedIn profiles and 952,242 experiences, collected between March and September 2026, almost all in France; four in ten belong to data occupations, two in three to tech occupations in the broad sense. This is not all of LinkedIn, of course; it is the population my searches target. Those whom no query has ever surfaced are therefore missing here too. The 20% of silent profiles is, moreover, a floor.
Reliability: an empty field in the database is an empty field on LinkedIn; 300 profiles checked by hand confirmed it. The void comes from the candidates, not from the collection.
Out of scope: ticked skills, whose collection is unstable from one crawl to the next; they will be measured separately.