AI Prompts at Work: Who Reads Your Logs?

Reading Time: 5 minutes

Your prompts show what you know, what you do not, and where you got stuck. So who else gets to read that diary?

AI prompts at work are now both a skill you build and a record someone else may keep.

🪙 The Token Diaries, Episode 4 of 8 — We have seen the developer’s ledger and the employer’s scoreboard. Today we switch sides and look at the employee: skill on one hand, paper trail on the other. New here? Start with Episode 1.

🪞 AI Prompts at Work: One Prompt, Two Meanings

Every time you ask an AI tool something, two things happen at once.

You get an answer. And you leave a record.

The answer builds your skill. The record is a log entry: what you asked, when, how you phrased it, and what you did next. For the employee, tokens cut both ways. Let us take the good side first.

🎓 The Skill Side: Fluency Pays

Knowing how to break a problem down, prompt well and check the output is becoming a hiring signal, not a hobby.

📌 Real-world case: Lightcast, which tracks job postings, analysed more than 1.3 billion of them in a report released in July 2025. Postings that list AI skills advertised salaries about 28% higher, nearly $18,000 a year, than those that did not. And 51% of postings requiring AI skills were outside IT and computer science. (Lightcast)

Read that with care. It measures advertised salaries in job postings, not what each person is actually paid. It also shows a trend: AI skill is spreading well beyond tech jobs.

The practical message is simple. The skill is portable. You take it to the next job, and nobody can revoke it with a policy change.

🪙 Token Truth: The skill is yours. The log is not always.

🧩 What “Fluency” Actually Means

It is easy to hear “AI skill” and think of clever prompts. That is the smaller half.

The bigger half is judgement. Can you break a messy problem into steps a model can handle? Can you tell when an answer sounds right but is wrong? Do you know which parts of the job should never be handed over?

Those are human skills with an AI interface. They are also the skills a prompt log reveals most honestly. Someone who asks sharp follow-up questions and checks the output shows a very different pattern from someone who pastes and ships.

That is why the same log that proves your skill can also expose your gaps. Which brings us to the question of who gets to read it.

🗃️ The Paper Trail Is Not Only Your Employer’s

When people worry about prompt logs, they usually picture a manager reading over their shoulder. The real picture is wider.

Your prompts sit on someone else’s servers. That means they can be pulled into disputes you have nothing to do with.

📌 Real-world case: In May 2025, a US court ordered OpenAI to keep records of its ChatGPT data in the copyright case brought by The New York Times. A judge lifted that blanket order in October 2025, but data already saved stayed accessible, and logs tied to accounts the Times flagged still had to be kept. Then, in a later order, the court told OpenAI to hand over roughly 20 million anonymised chat logs. OpenAI argued that 99.99 percent of the requested chats had nothing to do with the case, and it appealed. (Engadget, The Decoder)

Notice the detail that matters for employees. The logs were anonymised before being handed over. But “I deleted it” and “it no longer exists” are two different statements.

Remember Samsung from Episode 1, where engineers uploaded internal source code to ChatGPT. The lesson there was about where the data goes. The court case adds a second lesson: where it can be made to go.

🪙 Token Truth: Deleting a chat is not the same as erasing it.

🤫 The Shadow Side: What Employees Paste

Now flip the camera. If companies can see little of what employees do with AI, what are employees quietly doing?

📌 Real-world case: A 2026 survey by Wakefield Research for PagerDuty questioned 1,250 office professionals at companies with at least $500 million in revenue in Australia, Japan, the UK and the US, in April 2026. 34% said they had entered customer data into public AI tools. 31% said they had entered financial information or confidential company documents or strategies. And 66% had used AI at work even though they believed it was not allowed. (PagerDuty)

This is a vendor survey, so treat the percentages as directional. But the shape of the finding is hard to dismiss. Policies written for a world without AI are being quietly bypassed, and the risk lands on whoever pasted the data.

Seedhi baat: if your company has no approved AI tool, your people are already using unapproved ones. Banning does not remove the need. It removes the visibility.

🪙 Token Truth: Where the approved tool is missing, a shadow tool appears.

⚖️ When the Reader Is a Machine

There is one more twist, and it runs the other way. Employees are not only watched by systems. They are increasingly judged with the help of them.

📌 Real-world case: A ResumeBuilder survey of 1,342 managers, reported by People Matters in July 2025, found that 60% had consulted an AI model such as ChatGPT for guidance on layoffs, and nearly one in five said they let the tool make the final decision. It is a self-reported survey from a company that sells career products, so the numbers are indicative, not definitive. (People Matters)

Put the two halves together. Your prompts may be logged, and decisions about you may be shaped by someone else’s prompts. That is a lot of trust in tools whose rules you rarely see.

In Episode 1 we met the Amazon staff who told the Financial Times they suspected managers were watching their token counts anyway. The worry is not always about proof. It is about not knowing.

🪙 Token Truth: The scariest log is the one you do not know exists.

🏢 What a Good Employer Does Differently

Not every company handles this badly. The better ones tend to do four things.

They publish a plain-language AI policy that says which tools are approved. They explain what is logged, who can see it and how long it is kept. They separate usage analytics for teams from reading individual prompts. And they use the data to offer training and support, not to build a case.

None of this is expensive. It is mostly a matter of writing things down and sticking to them. The reward is the thing every AI rollout actually needs: employees who will tell you the truth about how they use the tools.

🛡️ An Employee’s Checklist

You do not need to fear AI at work. You need a few habits.

Ask what is logged. Which tool, which account, who can see it, how long it is kept.
Use the approved tool for work data. Keep personal accounts for personal things.
Do not paste what you could not email. Customer records, financials and credentials stay out.
Keep your own work log. A short note of what you shipped with AI and what it saved is your best evidence.
Learn the skill in the open. Fluency you can show beats fluency you hide.
Treat deletion as a courtesy, not a guarantee. Plan as if a log could outlive your chat window.

If you manage people, flip the list. Tell them what you log and why. Trust is cheaper than surveillance.

Sach batao: do you know what your company logs about your AI use, and who is allowed to read it?

🔜 Next Tuesday: Episode 5 — Selling by the Syllable

📬 Episode 5 is on its way. Subscribe to get it in your inbox the moment it goes live.

We have looked at the user, the developer and the employee. Next week we go to the other side of the counter: how AI vendors price, limit and profit from every token you spend.

Have a story about AI at work, good or bad? Drop it in the comments. The best ones may make it into the series.

📚 References

What’s your Reaction?