ScanMyJob: turning your work messages into a dated record of facts

You have two thousand messages with your manager. You have reread them, several times, late at night. And you still cannot choose between two readings that seem equally plausible: either something genuinely abnormal is happening, or you are reading far too much into ordinary clumsiness.

Nobody in that position is looking for software. They are looking for two very specific things, and those two things pull in opposite directions.

The first: to know whether they are overreacting. It is an honest question, and usually the one you cannot ask anyone — because asking someone close to you amounts to asking for confirmation, and people close to you almost always confirm. The second reading, the one that contradicts yours, never gets voiced. The second: to build a case. Here the need changes nature entirely. It is no longer about understanding, it is about producing — a document someone else can read, check and use.

These two needs call for different work, but they run into the same wall: two thousand messages cannot be reread. What is missing is not memory, it is ordering.

What people are actually searching for

Across a corpus of 2,545 real search queries about workplace difficulties, the most revealing phrasing is not the one you would expect.

It is not "am I being harassed at work".

It is "how to prove workplace harassment".

The distinction matters. People are not asking for an opinion, they are asking for a means. They already hold a conviction — right or wrong is not the point here — and what they lack is something to hand over. They know from experience that "my manager puts pressure on me" carries no weight, not with a lawyer, not with HR, not in a tribunal.

That is exactly the gap ScanMyJob sets out to fill.

What the tool does

You export your work conversations — a WhatsApp thread, a Slack channel, a Teams conversation, a run of emails — and import them. The service handles 2 to 15 participants: a one-to-one relationship, a small team, or an entire group channel.

You then declare who is who — which one is the manager, which is a peer, which comes from HR — and what you want from the report. You receive a full report, including a record of facts you can export separately.

One report, €8, one-off payment. No subscription, no cut-down version, no upper tier: there is a single report and it is the complete one. The price is charged in your own currency.

The number of participants is not a configuration detail: it changes what the material makes observable at all. With two people, you read a relationship — the sequence of exchanges, who follows up, who answers, how fast. With twelve, you read a structure: who sits at the centre of the information flow, who is kept outside it, who gets dropped from a thread and when. Those are not the same observations, so it is not the same report. And a measure that makes no sense in your configuration is not displayed as zero: it is marked not applicable. "Zero exclusions from threads" on a two-person conversation would read as an absence of fact when it is an absence of measurement — and in a document meant for a third party, that confusion is expensive.

The dated record of facts: the core of the product

This is the part to understand before any other, because it is what separates ScanMyJob from automated commentary.

An ordinary analysis writes: "communication appears strongly asymmetrical and intrusive". That may well be accurate. It is unusable.

A record of facts writes:

17 messages received after 8pm, between 3 March and 14 April. Six of them on a Saturday or Sunday. Median expected response time: under 40 minutes. Full list of the 17 messages, with date, time and sender.

Every line points back to a specific, dated, sourced message. Nothing is restated as a judgement. The document can be read end to end without anyone having to take your word for it: it offers something to verify, not something to endorse.

That is the form serious counterparties want. A lawyer does not build an argument on a feeling; they build it on exhibits. A labour inspector works from material evidence. An occupational health doctor needs temporal markers: since when, how often, with what progression. Three different uses of the same document, and all three assume dated facts.

The record exports separately from the report, precisely for that use: you attach the record to a file and keep the analysis for yourself.

The tool establishes facts, it does not issue a verdict

Here is the limit, and it is better read as a feature of the product than as a disclaimer.

ScanMyJob never states that a situation amounts to harassment. It does not say that a person has a disorder. It does not say you are burnt out. It does not say you are right.

In French law, workplace harassment is defined by article L1152-1 of the Labour Code, and qualifying a situation under it belongs to a judge — not to software, not to a report, not to a therapist. Other jurisdictions draw the line in their own way, but never in favour of an automated tool. Any product claiming otherwise would be doing its user no favours.

Why this is a strength rather than caution for show.

Picture the opposite report: a document concluding, in bold, that you are in a characterised harassment situation. You bring it to your lawyer. What happens next?

The first thing the other side does is attack the document. A piece of software issued a legal qualification from messages supplied by one party alone, without the other being heard, without context, without any adversarial process. The verdict collapses in one sentence — and it takes the surrounding facts down with it, now suspected of having been selected to produce it.

Someone who accuses on the strength of an automated verdict arrives weaker than they came. They traded solid material for an indefensible conclusion.

A record of facts does not fall the same way. The seventeen messages were sent or they were not. The dates check out. The qualification stays open — which is to say, it stays available to whoever is qualified to make it.

The report also keeps three registers strictly apart: facts, interpretations, hypotheses. You always know which one you are reading. And an observation that does not rest on enough material is not hedged — it is left out.

The three requests the tool refuses

They are stated up front rather than hidden, because an explained refusal beats disappointment after payment.

"Prove that I am being harassed." That is a request for a legal qualification. What the tool can do — and what serves that goal better — is establish the dated record of facts that will support it. A record can be put on the table. An automated verdict cannot. "Prove that I am right." A tool that promises to prove you right has stopped measuring anything. All it can produce is confirmation, and confirmation teaches its reader nothing — least of all what they needed to know before acting. "Tell me whether my manager is a narcissist." That request pins a clinical label on a person who was never heard, never consented to being assessed, and of whom nothing is known beyond what they wrote in a professional setting. No serious practitioner makes a diagnosis under those conditions; software still less.

The three refusals have one thing in common: they protect the user from their own document.

Names never leave your browser

At import, full names, phone numbers and addresses are replaced on your own device, before anything is sent. No real name reaches a server.

What replaces them is not an anonymous serial number but initials and a role:

```
John Doherty → J.D. (manager)
Sarah Mitchell → S.M. (colleague)
```

The difference with an anonymous "Person G" looks cosmetic. It is decisive.

You know your own team. Nobody is protected by hiding from you who is who in your own conversation — it only stops you using the document. A report reading "Person G copied Person D on Person B's reply" becomes unreadable within three lines, and unreadable means unusable: neither you, nor your lawyer, nor the occupational health doctor will get to the end of it.

"J.D. (manager) copied you in" is understood instantly. And yet the document itself names nobody: it cannot circulate as a named accusation.

Two honest qualifications, because they matter.

The role is declared by you, never guessed. The tool does not try to infer who the manager is from the tone of the messages — that would be a source of error, and a silent one. You declare it at import. Initials are not absolute anonymity. In a team of twelve, "J.D. (manager)" remains identifiable to anyone who knows the team. This is pseudonymisation, not disappearance. The real safeguard sits elsewhere: the document establishes facts and passes no judgement on J.D. A dated fact attributed to J.D. can be put forward; an opinion about J.D. could be challenged. And reintroducing full names before passing the document on is your decision alone, made knowingly.

What you ask the report for: eight objectives

Before payment, on the same screen where you declare the roles, you tick what you want. Three objectives at most, and the order counts: the first one drives how the report opens.

Understand what is going on — the observed dynamic, with no starting assumption.

Find out whether I am reading too much into it

Document dated facts

Decide: stay or leave — psychological contract, organisational justice, person-job fit; the report closes on what is reversible and what is not.

Prepare for a meeting — chronology of friction points, phrasings on record, a preparation grid.

Prepare for a negotiation — the margins visible in the exchanges, fallback position, anchors.

Bring the tension down — conflict-handling modes, escalation stage, and what has already defused things in your own exchanges.

Understand my own part in it

Three of these deserve unpacking, because they carry most of the value.

"Find out whether I am reading too much into it"

This objective does something counter-intuitive: it foregrounds what contradicts your reading as much as what supports it.

In practice, the section gathering the material that cuts against your interpretation does not stay as a footnote at the end — it becomes a full section, near the top of the report.

It is the most uncomfortable of the eight and the most useful. Someone who discovers that twelve of the twenty things worrying them are explained by a context they had not connected, and that eight are not, walks away with something solid. They know which eight.

It is also the one a commercially minded tool does not dare offer, because it disappoints a share of its buyers. Which is exactly why it is worth something.

"Document dated facts"

Tick this one first and the factual record stops being an appendix: it moves to the front and becomes the main object of the document. The psychological analysis moves behind it. The separate export is put forward.

This is the objective to choose if your need is to build a case — and it is the one that genuinely answers "how to prove", far better than a peremptory conclusion would.

"Understand my own part in it"

What you do that keeps the situation running. Phrasings that close an exchange down, follow-ups that produce the opposite of their intent, silences read as refusals, requests worded so indirectly that no answer was possible.

This is not an invitation to take responsibility for what is being done to you: genuinely being on the receiving end of something is not "playing the victim", and the tool is explicitly built not to conflate the two. It is simply the only lever entirely within your reach. On everything else, you depend on how someone else reacts.

Two safeguards apply here as elsewhere. An objective cannot make the report say what the data does not carry: ticking "prepare for a negotiation" on material containing no trace of one does not manufacture a section — the report states that the question is not documented in the exchanges provided. And some objectives conflict with the configuration: "understand my own part in it" on a fifteen-person channel yields little, because at that scale you are no longer observing a relationship. The report says so rather than padding.

49 models, never all at once

The analysis draws on 49 models from work psychology, social psychology and negotiation research: the job demands-resources model (JD-R), Karasek, Siegrist, the psychological contract, organisational justice, psychological safety, leader-member exchange (LMX), the Thomas-Kilmann conflict modes, the Harvard negotiation method, social network analysis, attribution biases, and others.

The important point lies elsewhere: they are never all applied together. That would be the opposite of analysis — stack enough grids on any material and something always comes out, and that something means nothing.

Models are selected according to the situation. An exchange with your direct manager does not call for the same frameworks as a twelve-person team channel: the first draws on leader-member exchange, the psychological contract, psychological safety; the second draws on structural analysis, decision rights, information flow — frameworks that are strictly inert on a two-person conversation.

And no significant conclusion rests on a single model: at least two, converging, are required before an observation goes into the report. Each indicator also carries its false friends — the things that look like it without being it. Being copied in is not being consulted. A sign-off is not a decision. A silence is not a refusal. That is written into the models themselves, precisely so the report does not conclude from a resemblance.

Key takeaways

What ScanMyJob produces: a report from your real work conversations, 2 to 15 participants, including a dated, sourced, exportable record of facts. €8, once. Where the value sits: the move from impression to count. "He puts pressure on me" becomes "17 follow-ups after 8pm over 6 weeks", with the list attached. What it refuses: to qualify a situation in law, to prove you right, to pin a clinical label on an absent third party. Those three refusals are not limitations endured — they are what makes the document defensible. What stays with you: the names. They are replaced by initials and a role before anything is sent, in a form that keeps the document readable for those who will need to use it.

And the question that remains yours, the one no report will settle: what you decide to do with it.

Explore ScanMyJob →

---

FAQ

Can the report be used in an employment tribunal?

The record of facts is designed to be attached to a file: it is dated, sourced, and every item points back to a specific message. But it is a lawyer who decides what is submitted and how it is presented — the tool produces an exhibit, it does not build a strategy. One useful note: case law on the admissibility of evidence in employment matters has been moving in recent years, and that question belongs with a professional, not with an article.

What if my messages show nothing in particular?

The report says so. That is one of the answers it is built to be able to give, and it is what the "am I reading too much into it" objective guarantees. A tool incapable of concluding that there is no case would carry no informative value at all — it would only hand each user back what they brought.

My colleagues never agreed to be analysed. Is that a problem?

That is precisely why anonymisation happens at import, on your device, before anything is sent. People mentioned are designated only by initials and role, and the report passes no judgement on them. That reduces the risk without eliminating it: in a small team, initials remain recognisable. What you then do with the document is on you.

Is there a minimum number of messages?

No threshold is stated, but the logic is straightforward: the record measures repetitions, frequencies and delays. On a handful of messages there is nothing to count. This is a tool that makes sense over weeks or months of exchanges, not over one bad day.

How is this different from asking a general-purpose chatbot?

Two practical differences. First, the ordering: the record counts, dates and places each occurrence within a sequence, where a conversational answer summarises and smooths. Second, the framing: a general-purpose assistant will happily qualify, diagnose and tell you that you are right — that is in fact what it does most readily. Here those outputs are ruled out by design, which is less pleasant and considerably more useful if the document is going to be read by someone other than you.