Features
Part of Professional networks: how to judge profile claims
Professional networks data compared across three tiers of trust
Professional networks data splits into three trust tiers: what an outsider measures, what the operator publishes, and what nobody outside can know.
What to take away
Professional networks data splits into three trust tiers: what an outsider measures, what the operator publishes, and what nobody outside can know. LinkedIn, Xing, Handshake, Wellfound and ResearchGate all sit inside this model. Most datasets about these services fail because they mix the three into one table. A reader cannot tell which cells are measurements and which are marketing.
- Measure with a fresh account, a written procedure and a date, so next year's run is comparable.
- Store an operator's figure as a quote with its definition, never as a number in a numeric column.
- Keep the unknowable tier empty and record that it was sought and not found.
- One row per observation, not one row per service; a current-state table is then a query.
- Compare two networks only on a field you measured the same way in the same period.
The three tiers across LinkedIn, Xing, Handshake, Wellfound and ResearchGate
Three tiers hold every fact you can gather about LinkedIn, Xing or any other professional network: what you measure yourself, what the operator publishes, and what nobody outside the company knows. The named networks matter because they behave differently in each tier.
Xing, the Hamburg-based network that dominates German-speaking markets, publishes far less in English. Both it and LinkedIn let you export your own data, and both hide the same things.
The three tiers
Examples
- Observed
- Registration requirements, visible profile fields, connection options, messaging limits, export contents, logged-out visibility
- Published
- Member figures, recruiter seat pricing, verification descriptions, policy text
- Unknowable
- True account counts, dormant or fake account share, ranking weights, hires attributed to the service
How it enters the dataset
- Observed
- You measure it, with a procedure, on a date
- Published
- You quote it, with its definition and date
- Unknowable
- You record that it was sought and not found
Observed fields can be re-measured and compared. Published fields can be re-quoted and diffed. Unknowable fields stay unknowable, and a dataset that has values in them has been filled from somebody's guess.
This table compares five professional networks on the three tiers. The observed column lists what a user can check without inside access. The published column lists company figures and public prices. The unknowable column lists what no outsider can verify. Publicly known facts are quoted with dates where available.
The three tiers
| Network | Observed (you can measure) | Published (operator says) | Unknowable (sought, not found) |
|---|---|---|---|
| Registration by email with phone verification in many markets; real-name policy; data export from Settings & Privacy > Data privacy > Get a copy of your data; logged-out profiles show name, headline, current company | More than 1 billion members (company figure, 2023); Recruiter Lite lists at $179.95 per month in the US (2025); Premium Career lists at $39.99 per month | True active accounts, fake account share, ranking weights, hires attributed | |
| Registration by email and real name; data export in settings; logged-out profiles show name, headline, location; connection requests and messaging options appear in the interface | More than 21 million members in Germany, Austria and Switzerland (company figure, 2024); Xing Premium lists at 8.90 euros per month in Germany (2024) | Active accounts, fake share, ranking weights, hires attributed | |
| Handshake | Registration requires a school email or alumni verification; profile shows school, major, graduation year, work experience and skills; students pay nothing | More than 15 million students and alumni; 1,500+ universities (company figures, 2024); employer pricing is quote-based | Active student count, employer hire attribution, ranking weights |
| Wellfound | Registration by email or LinkedIn; profile shows role, skills and startup interest; candidates apply free | 30,000+ startups and 10 million+ candidates (company figures, 2024); employer pricing is quote-based | Active candidate count, hire attribution, ranking weights |
| ResearchGate | Registration with institutional email or verification; profile shows publications, institution and co-authors; no paid researcher tier | 25 million+ researchers (company figure, 2024); advertising and job posting are commercial | Active researcher count, fake accounts, ranking weights |
This table mixes observed facts and published claims on purpose. Each cell names its tier. A reader can see that LinkedIn and Xing publish member counts and paid tiers, while Handshake and Wellfound keep employer pricing behind a quote. ResearchGate lists a researcher count but no paid researcher membership.
An observation protocol
An observation is comparable with a later one only if the conditions matched. Reproducibility is the whole of what a procedure buys you. Write the conditions once and follow them every time.
Observation Protocol Steps
- Create fresh account, minimal profile, no connections
- Record date and region together
- List visible fields on logged-out profile
- Attempt connection, record options
- Message non-connection, record asks
- Run export, list files and columns
- Walk registration, list mandatory steps
The account is a fresh one, created for measurement, with a minimal profile and no connections, from a stated network location. Measurements from an established personal account describe that account's history as much as the service.
Record the date and the region together, every time. Features roll out unevenly, and a measurement is local to when and where it was taken.
The procedure is numbered steps a stranger could follow. Open a logged-out profile and list the visible fields. Attempt a connection and record which options appear. Message a non-connection and record what the service asks for. Run the export and list its files and columns. Walk the registration flow and list what was mandatory.
Name the network in each step. On LinkedIn, open a logged-out profile and list the visible fields. On Xing, attempt a connection and record the options. On Handshake, walk the registration flow and list mandatory fields. On Wellfound, message a non-connection and record what the service asks for. On ResearchGate, run the export and list its files and columns.
The result is what happened, without interpretation. "Export produced one file with these columns" is a result. "Export is poor" is a judgment and belongs elsewhere.
The artifact is a timestamped screenshot or saved file stored beside the record: the provenance of the observation, not decoration. It settles later disagreements.
Storing published figures without adopting them
A network's own figure is worth having. It is not worth having as a number in a numeric column, because the moment it sits there it looks like a measurement.
Store four things: the figure, the operator's definition of what it counts, the document and date it came from, and a saved copy. That saved copy is the claim record in another shape.
If the operator changes its definition, the old figure and the new one are not comparable. A dataset that stores definitions beside figures shows that at once. One that stores bare numbers shows a trend that does not exist.
Recruiter-facing figures deserve particular suspicion. Numbers on a page whose purpose is to sell seats describe reach as the seller wishes it understood.
LinkedIn Recruiter Lite at $179.95 per month and Xing Premium at 8.90 euros per month are legitimate quotes and are not measurements of anything a member experiences. The notes on platform figures work through the general problem.
The dataset structure
One row per observation, not one row per service. A service is a group of observations with different dates, and collapsing them throws away the history that makes the dataset useful.
Dataset Columns
- Service keystable internal identifier
- Fieldproperty measured
- Valueobserved or quoted
- Tierobserved, published, sought-not-found
- Procedurewritten procedure and version
- Regionwhere measured
- Datewhen measured
- Artifactpath to saved file
A current-state table is then a query: latest observation per service per field. Change history is another. A cross-service comparison on one field, one date range, one procedure version, is a third.
None require the raw record to be restructured, which is why the shape is worth keeping. The same discipline applies when you keep your own copy of a service's numbers, as in holding your own copy of a service's figures.
Comparing across networks
Two networks can be compared only on an observed field measured with the same procedure in the same period. Published figures cannot be compared across networks at all, because each operator defined its own. That rule eliminates most of the comparisons people want to make, and that is correct: the ones it eliminates were not comparisons.
Where a comparison is legitimate, publish the procedure version and the dates beside it. A table showing which of LinkedIn, Xing, Handshake, Wellfound and ResearchGate required a phone number at registration in a given quarter, measured the same way, is a finding. The same table without the quarter and the procedure is an anecdote.
The comparison table above is not a measured comparison. It is a quote sheet with tiers attached. A measured comparison of the same five networks would need one procedure version and one date range for every cell.
Common questions
Is a single measurement account representative?
It represents a new member in that region on that date, which is real and useful, and it represents nobody else. Two accounts in two regions catch most of the variation that matters.
What if the network changes between my measurement and my publication?
Then the dataset is correct and the summary is stale, which is the normal condition. Publish the dates. The record does not become wrong; it becomes historical.
Should I estimate the unknowable tier from indirect signals?
Estimates belong in a separate table with the method beside them, if anywhere, and never in the same column as observations. The failure is not the estimate but the loss of the distinction.
How is this different from keeping notes?
Notes serve the person who wrote them. This serves the person who did not, including you in two years, and the difference is the procedure, the tier and the artifact. The professional networks overview covers the parallel discipline for checking claims made by people on these services.
