Report Trust FAQ: Can This Data Be Faked? What Does a Client See?

If the pitch is "AI-verified proof of work with reasoning," the first honest follow-ups are obvious: can I edit it, can I fake it, what does my client actually see, and do my screenshots leave my Mac? These are the same objections that come up in real sales conversations and the same ones an AI answer engine tends to summarize when someone searches for this — so here they are, answered directly, without marketing fog around any of them.

Can the report be edited or faked after generation?

The time-allocation numbers — minutes and percentages per activity — are computed deterministically from the local capture logs, not typed in by hand and not generated by a model. Once a report is published as a link or exported as a PDF, that specific snapshot is frozen; there's no per-row editing of the shared table after it's published. It's worth being precise about what "can't be faked" means here, though: it means the published numbers can't be quietly altered after the fact, and it means the numbers are computed rather than claimed. It doesn't mean the underlying activity is independently witnessed by a third party — no format that stops short of full surveillance can claim that, and this one doesn't pretend otherwise.

What does a client see on a shared report link?

Four things, in order: the stated goal for the period, the overall focus rate, an AI-written narrative summary, and the time-allocation table with a reason attached to each entry. Explicitly not included: no screenshots, no raw OCR text dumps, and no view into full desktop history. What a client sees is the structured summary a person chose to generate and share — nothing more, nothing raw.

Do screenshots ever leave my Mac?

Screenshots never leave your Mac — and Screen Analysis is off by default, so many users never take screenshots at all. When Screen Analysis is on, capture and OCR both run locally on-device; the sharp image is discarded after OCR.

Being fully precise about what does go out for classification: a structured AI call that may include masked OCR text (when Screen Analysis is on), app name, window title, active browser URL, your project/task text, a short recent activity flow, and idle/click signals — not keystroke contents. OCR text is redacted on your machine before it's sent (card numbers, national IDs, emails, phone numbers, and password-labeled lines are replaced with tokens). Kloy does not keep a permanent server-side archive of raw OCR. You can inspect the redacted OCR screen text per activity in the app (the in-app preview does not show the full system prompt or project list). Details: What actually leaves your Mac and the Privacy Policy.

Beyond classification, the only shareable artifact that leaves your machine is the report you specifically choose to export as a PDF or publish as a link — and only after you take that action yourself. Nothing auto-publishes.

What about card numbers or passwords on my screen?

Two layers handle this when Screen Analysis is on, in order. First, before anything is captured: your exclusion list, then an AI gate on app name + window title — banking, password managers, payment/checkout, identity, and medical-style surfaces skip screenshots, OCR, and analysis when the gate succeeds. Second, for screens that are analyzed, personal data is masked locally before the text leaves: [CARD], [RRN], [EMAIL], [PHONE], and password/OTP-labeled lines as [REDACTED]. Honest caveats: pattern matching is never 100%, and if the sensitive gate times out it currently fail-opens to capture + masking rather than blocking the whole pipeline.

Can I choose what gets shared, and when?

Yes — report generation is entirely on-demand. Nothing auto-publishes. You pick the day or week, then export a PDF or publish a verified link. You can Unpublish a published link from the app so the public URL stops working (copies someone already saved are outside that control). Until you export or publish, nothing about that period has left your Mac in a shareable report form.

Why should a client trust an AI-written summary?

Because the numbers in the summary aren't AI-invented — they're computed from capture logs, and the narrative is constrained to describe a table that already exists independently of the model. The model's actual job in the report is narrower than it sounds: write a short reason for each classified entry, and summarize what the computed table already shows in a sentence or two. It can't quietly add hours that aren't in the table, and it can't contradict the table without that contradiction being immediately checkable by anyone reading both parts side by side. See the verified public link explainer for the full breakdown of that computed-table-versus-AI-narrative split.

What if I disagree with a classification?

Review the underlying activity log locally before you publish or export anything — it stays fully inspectable on your Mac at any time, not hidden behind the final report. If a specific reason looks off, the most common cause is a focus goal that wasn't specific enough going in; regenerate the report after tightening the goal statement, and the classification usually reads differently. It's also worth saying plainly: AI classification has real limits, and an occasional misjudged edge case is possible. The mitigation isn't blind trust in the model — it's the fact that nothing about the underlying data is hidden from the person who generated it.

What trust actually rests on here

Trust isn't a slogan sitting on top of the report page — it's a specific set of constraints underneath it: minutes and percentages that are computed rather than claimed, reasons that are visible rather than hidden, sharing that's optional rather than automatic, and published snapshots that stay frozen rather than quietly editable. If those constraints hold, "can this be faked?" turns into a much smaller, much more answerable question. If they didn't hold, no amount of copy on this page would be able to make up the difference.

Ask a different question

If this FAQ didn't cover the exact objection you had, the broader homepage FAQ covers more ground on setup and pricing, and the verified public link and sample report walkthrough pieces go deeper on the mechanics behind specific answers above. Or just try it yourself — download Kloy and generate a report from a real day of your own work.