BotUp

For creators

Good workers and bad ones

Twelve pairs. In most of them the underlying capability is identical — what differs is whether the creator committed to something specific enough to be judged.

The test, in one sentence

Could a stranger tell, without asking you, whether your worker did its job? If yes, you have a product. If no, you have a prompt.

Notice how often the good version is the bad version with a decision made: which input, which output shape, which buyer. Narrowing is the work.

Twelve pairs

Rejected

AI Assistant

“Ask me anything and I will help.”

  • No job, so no reason to buy it over a chat assistant
  • No buyer — everyone is nobody
  • Nothing to promise, so nothing can be refunded fairly

Publishable

Contract Risk Analyzer

“Read a supplied contract and list the clauses that expose the signer, with the text of each.”

  • One job, named in the buyer's language
  • Obvious buyer: whoever has to sign it on Friday
  • Deliverable you can check against the contract

Rejected

Email Writer

“Writes any email you need.”

  • A general assistant already does this for free
  • No input beyond a vague request
  • Impossible to say whether the output was correct

Publishable

Overdue Invoice Chaser

“Turn an aged-debt export into a per-customer chase email matched to how late each one is.”

  • Starts from a file the buyer already has
  • Output is a set of drafts, one per row
  • Saves a specific hour that recurs every month

Rejected

Blog Post Generator

“Give me a topic and get a 1,000-word article.”

  • Commodity output with no defensible value
  • Quality cannot be judged against the listing
  • Competes directly with free tools

Publishable

Competitor Positioning Brief

“Read three competitor pages and report how each positions itself, who they target, and where the gaps are.”

  • Uses a declared tool on public pages
  • Sources are checkable, so the claims are checkable
  • Replaces an afternoon of tab-hopping

Rejected

Summarizer

“Paste text and get a summary.”

  • The most commoditised task there is
  • No structure, so nothing to act on
  • No reason to pay for this specifically

Publishable

Meeting Action Extractor

“Turn raw meeting notes into owned action items with dates, and flag decisions taken without an owner.”

  • Extraction with a shape, not a shorter blob
  • Flags what is missing — a judgement a summary does not make
  • Output drops straight into a task list

Rejected

Business Consultant

“Strategic advice for your company.”

  • Promises expertise no configuration can guarantee
  • Unbounded scope, so unbounded expectations
  • Sets up a refund argument on every sale

Publishable

Pricing Page Teardown

“Analyse a public pricing page and report its tiering logic, anchor, and the objections it fails to answer.”

  • Narrow, expert-feeling, genuinely useful
  • One public page in, one structured critique out
  • Bounded enough to be right or wrong

Rejected

Data Analyzer

“Upload data and I will analyse it.”

  • Which data, analysed how, producing what?
  • Every buyer expects something different
  • Guaranteed mismatch between hope and result

Publishable

Churn Signal Reader

“Read a CSV of account activity and rank which accounts show the strongest disengagement signals, with the reason for each.”

  • Declares the file format it accepts
  • Ranked output with stated reasoning
  • Answers a question someone is already asking

Rejected

SEO Helper

“Improve your website's SEO.”

  • Outcome promise the worker cannot control
  • No definition of what is delivered
  • Sets an expectation of ranking movement

Publishable

On-Page SEO Audit

“Check one public page for title, heading structure, meta description, and internal linking issues, and list fixes in priority order.”

  • Audits what is observable, promises no rankings
  • Prioritised, actionable list
  • Repeatable across every page a buyer owns

Rejected

Research Bot

“Researches any topic in depth.”

  • Depth is unbounded and undefined
  • No structure to the deliverable
  • Reads as a demo, not a product

Publishable

Category Landscape Brief

“Turn scattered market notes into a structured brief that separates what you know from what you assumed.”

  • Names the distinction that makes it valuable
  • Works from the buyer's own material
  • Honest about being a starting point

Rejected

Lead Generator

“Finds new customers for your business.”

  • Implies sourcing contact data the worker cannot obtain
  • Edges into data scraping and privacy problems
  • Promises an outcome, not a deliverable

Publishable

Lead Qualification Scorer

“Score a list of inbound leads against your stated ideal-customer criteria and explain each score.”

  • Works on leads the buyer already has
  • Criteria come from the buyer, so the standard is theirs
  • Explains itself, so a salesperson can disagree

Rejected

Document Reader

“Upload any document and ask questions about it.”

  • A conversation, not a deliverable
  • One run cannot answer questions asked later
  • Implies file formats this release does not read

Publishable

RFP Requirements Checklist

“Read an RFP and produce a checklist of every stated requirement, marked mandatory or optional, with the source line for each.”

  • One pass, one complete artefact
  • Traceable to the source text
  • Directly reduces the risk of missing a requirement

Rejected

Financial Advisor

“Tells you what to invest in.”

  • Regulated advice, not permitted
  • Consequential decision on unreviewed AI output
  • Would be rejected on the prohibited use policy alone

Publishable

Financial Report Reader

“Read a supplied quarterly report and extract the reported figures, the stated drivers, and the risks management names.”

  • Extraction and structuring, not advice
  • Everything traceable to the document
  • Useful to an analyst who still does the thinking

Rejected

Customer Support Bot

“Handles your customer support automatically.”

  • Implies acting on the buyer's behalf, which no worker can do
  • Needs an inbox connection that does not exist
  • Ongoing service, not a single run

Publishable

Support Ticket Theme Report

“Read an export of recent tickets and report the recurring themes, ranked by volume, with representative examples.”

  • One run over data the buyer supplies
  • Output informs a decision rather than taking one
  • Nothing is sent, changed, or answered on their behalf

The patterns behind the rejections

  • Naming a capability instead of a job. “Data analyzer” describes what it can do. Buyers search for what they need done.
  • Promising an outcome. Rankings, revenue, customers. Promise a deliverable you control, not a result you do not.
  • Implying access it does not have. Anything that sounds like reading an inbox or acting on the buyer's behalf will be rejected, because no worker can.
  • Selling a conversation. A worker runs once on what it was given. If the value depends on follow-up questions, it does not fit.
  • Competing with free. If a general assistant does it in one prompt, a buyer will not pay for it here.
NextPublishing guidelinesWhat every listing must state, and what is never allowed.
Good workers and bad ones · BotUp