Skip to content

How we label cases

Every case is a public report from someone who says Grok Bot did a household money or admin job. The savings are what they report; nothing here is verified with the provider. What we can do is say plainly what each source shows, and keep the claim separate from our judgment of it.

Where the cases come from

So far all 143 cases are public posts on X, collected from threads and searches in October 2026, and the labels were last checked on 2026-10-20. Each case links to its source and credits its author. The claim, the ask, and the steps are paraphrased from the post and are never edited to make a case look stronger. We link to posts instead of copying their screenshots, and we never show order, confirmation, or account numbers.

The pool is selected: people post what worked. So the share of "done" cases tells you nothing about how often the bot succeeds.

Evidence: what the post actually shows

The strongest thing visible in the post, judged by opening the attached images.

Evidence: Text only 123 cases
The post states the result but shows nothing that backs it up.
Evidence: Bot's own report 15 cases
The post shows the bot's own messages. The provider's side is not shown.
Evidence: Provider confirmation shown 5 cases
The post shows the provider's own email, chat or confirmation page.
Evidence: Settled on a bill or statement 0 cases
The post shows a bank line, a posted refund, or a later bill with the change.

A screenshot of the bot quoting a confirmation number or a price is still the bot's own report. When a provider's page shows up inside a bot screenshot, it counts as provider confirmation, but often only for the order or refund, not the price. Those cases carry a one-line note on what the provider's side actually shows, and their amounts are labeled by what backs the number itself.

Stage: how far it got

Stage: Reported done 102
As reported, the change is in effect or the money has landed. This says what was reported, not how well it is proven.
Stage: Provider agreed 7
The provider accepted, but the change isn't in effect or the money hasn't landed yet.
Stage: Waiting on provider 1
Waiting on the provider when the post went up.
Stage: Request sent 13
A request, claim or opt-out was sent. No answer reported.
Stage: Found, not requested 11
The bot found an opportunity, but nothing was requested.
Stage: Nothing to change 2
The check ran and found nothing to change.
Stage: Didn't work 4
The attempt did not work.
Stage: Stage unclear 3
The post doesn't make the outcome clear.

Read the stage together with the evidence. "Reported done" with "Text only" is a claim and nothing more.

How cases are grouped and sorted

Cases are grouped by outcome, and the biggest saving comes first. Proof is a label on every case and a filter, not a gate: a text-only claim of a big saving is shown, marked as text only. Use the proof filter or sort to see the better-backed cases first. Unfinished, failed and flagged cases are one click away and visibly marked, never hidden.

  • Reported done (102)

    The provider agreed or the change is reported done. Biggest saving first. Each case says what backs it.

  • Not finished yet (25)

    The bot found something, sent a request, or is waiting on the provider. No outcome yet.

  • Didn't work, or nothing to change (6)

    The attempt was blocked or rejected, or the check found nothing to change. Useful for knowing the limits.

  • Flagged cases (10)

    The author promotes a related product, the story is doubted, the money is business spending, or another tool may have done the work.

Caveats

  • Author promotes a related product. The author sells or promotes a product, course or bot tied to the post. Moves a case to "Flagged cases" and out of totals.
  • Publicly doubted. There is credible public doubt that the story happened. Moves a case to "Flagged cases" and out of totals.
  • Business spending, not household. The money is business or developer spending, or rental property. Moves a case to "Flagged cases" and out of totals.
  • Another tool may have done it. Another product may have done the work. Moves a case to "Flagged cases" and out of totals.
  • Numbers don't agree. The post's numbers don't agree with each other or with its own screenshot.
  • May have happened without the bot. The result may have happened without the bot.

How money is counted

  • Amounts are the poster's own numbers, in the currency they used. Every amount says what backs it: the poster's claim, the bot's own report, or something shown by the provider. When the provider's page confirms the order but not the price, the amount is labeled as the bot's report.
  • The rough first-year value adds a yearly amount (or 12 times a monthly one) to any one-time amount, converted to USD at ECB reference rates for 2026-10-19. Each case says how its value was built.
  • Totals only count cases where the provider agreed or the change is reported done. Potential savings (a quote or a finding not acted on) and flagged cases are left out. Totals are split by what backs each amount.
  • No total on this site is "confirmed savings". A cheaper plan may be slower, a lower price may be a 12-month promo, and a refund may be promised but not yet paid. Individual cases matter more than any total.

Prompts

On a case page, "Prompt (verbatim)" is quoted from the post and "Prompt (adapted)" is a paraphrase. The prompts on job pages are written for this site from what the posts did, and are always labeled adapted.

Every job prompt builds in a step where you approve before anything is accepted, paid, or sent. That pattern is also used in the seven bill-cutting prompts AI Guides shared on X.

Tested by VelvetShark

Some jobs are also run from Poland for the VelvetShark video. Those results show up on the case page once they are in. Not having a US account is a limit of that testing, not evidence that the job can't be done.

Corrections and removal

If you posted one of these cases and want it corrected or removed, or you have the provider's confirmation for a claim, message @velvet_shark on X.

Not affiliated with xAI. Grok is a trademark of its owner.