Hive Asset Group, LLC
A Minnesota company. Paperboy is one of its products, built by a former law enforcement officer turned founder.
About
Articles built to rank, with every claim checked before they go live
Paperboy is a product of Hive Asset Group, LLC, a Minnesota company. It is built by a former law enforcement officer turned founder. Here is who builds it, how it works and what the fact check does.
- No card for the first article
- Open source under AGPL-3.0
- Every number tied to a saved source
Draft, 3 claims
Click a sentence to see its source.
support.google.com Verified
Search Console shows you data for the past 16 months.
Number and wording match the source.
Example text and sources, written for this demo.
- 20articles the engine published on its own blog
- 120sources saved and checked against
- 1articles held back by the fact check until fixed
Live from Paperboy's own database. Paperboy runs on its own blog. See verified results
Why it exists
Why does Paperboy exist?
A small business site needs new articles every week, and the owner has a business to run.
Most AI writing tools solve the writing and leave the checking to you. So the work you wanted off your plate comes back as a pile of drafts to verify by hand.
We built Paperboy to do it the other way round. It checks first and publishes second, and it holds an article back when it cannot back up what it says. We run it on our own blog first, so we feel every miss before you do.
See verified resultsThe method
How does the fact check work?
The check is code first and a model second. It runs on every article, on every plan.
- 1
Numbers
Every number in a cited sentence must literally appear in the text of the source that sentence links to. A shown calculation, such as "$85 times 12 equals $1,020", is checked through its inputs.
- 2
Links
Every link must point at a source that was actually read for this article.
- 3
Sources that sell something
If a cited source sells a related product, the article has to say so.
- 4
Citations
A model reads each cited sentence next to the passages of its source most likely to support it, and answers supported, partly supported or not supported. A "not supported" verdict gets a second look with more of the source before it can block publishing.
- 5
Voice and claims
A second model pass checks the article against the customer's voice, flags specific claims with no source, and catches any claim the customer banned.
- 6
The verdict
An article passes with no blocking issue and a score of 70 or more. If it fails, only the flagged paragraphs are fixed, once, and it is checked again. If it still fails, it is marked "Needs your review" and not published.
What the check does not do
It checks an article against its sources. It cannot prove a source is right. It cannot know your pricing change from yesterday, so use Approve first for anything about your own product.
You see the full fact check for every article.
From our own blog, not a demo
Does the check really catch things?
These are real fact checks from paperboyseo.com. One was published. One was held back until it was fixed.
Verified Published
36 of 36
cited claims checked against the saved source text.
- Every claim has a source: passed
- Numbers match their sources: passedEvery number in a cited sentence was looked up in the saved source text.
- Sources say what the article says: passedEach cited sentence was compared with the passages of its source.
- Sounds like the customer: passedVoice match 92 of 100.
- No banned claims: passed
- House style: passed
Held Caught before it went live
Fact check
RightBlogger alternatives: 6 options compared
Not published1 issue
to fix before this article can publish. The engine held it back.
- Every claim has a source: passed
- Numbers match their sources: held1 item held for a fix
- Sources say what the article says: passedEach cited sentence was compared with the passages of its source.
- Sounds like the customer: passedVoice match 92 of 100.
- No banned claims: passed
- House style: passed
Under the hood
How is Paperboy built?
The product
A Next.js app on Netlify, with accounts and data in Supabase (Postgres) and payments through Stripe.
The writing
Claude models from Anthropic. Claude Sonnet 5.5 writes and does the one rewrite. Claude Haiku 4.5 picks topics, searches, reads sources and judges the fact check.
The research
Up to 6 web searches per article. We open each source ourselves and save its text, so the fact check compares the article with the real page and not a search snippet.
The limits
One article may use at most $1.50 of AI cost. If a run would cross that, it stops and the article is held for you.
Publishing
A blog hosted on your own subdomain, WordPress, Ghost, a webhook, or a prompt you paste into a site builder. Approve first, or leave it on autopilot.
Billing and data
Stripe handles cards, and we never see a full card number. What we store, and how to delete it, is on the security page.
Open source
Is Paperboy open source?
The loop is. Topics, research, writing, publishing and the rank loop are AGPL-3.0, and the public repo is coming soon. The tuned fact-check verifier, cross-site learning and billing stay closed.
Run it yourself
Self-hosting for your own sites is free. You bring your keys and pay your own model bill. The public repo is coming soon (v0.1 alpha).
Read the self-host guideOr let us run it
The cloud adds the tuned verifier, hosted blogs, the dashboard and calendar, and the MCP server. Your first article is free.
See pricingThe company
Who is behind Paperboy?
More from Hive Asset Group, LLC
- Riot Digital riotdigitalai.com
- PickCheck pickcheck.co
- Stash getstash.cloud
Next
How do I get in touch?
Email hello@paperboyseo.com, or pick a page below.
Read one before you trust it
Paste your address and read a finished article with its full fact check. Free, no card.
Prefer to run it yourself? Self-host it (repo coming soon). Using an agent? Add the MCP server.