No-AI Standard

Human Research Without AI Shortcuts

Research is where the temptation sits. A writer can quietly ask for facts, but not draft with a model. This policy aims to close that gap and states how we verify those citations.

Order content How it works

What the No AI Research Policy Requires

This rule has a narrow scope. No generative model may be asked to provide a fact, a figure, a source, a definition, a summary, or a quote, all of which may contribute to the shaping of an article. It covers the apparent cases and the more subtle cases, including the case of inserting a report into a chat window and listening to the response rather than reading the report.

A journalist has a similar work style to a writer. They open the study, filing, manual, or transcript; read to the side of the sentence; and understand what is needed. If a subject requires a person, a writer emails or calls. If there is a paywall, we purchase a subscription.

There is one fair difficulty. Search results come with a generated summary positioned above the links, and a writer is restricted in their ability to turn it off. Our position is that the summary is a signpost and never a source. If a claim exists solely within that box, the claim is omitted from the draft until the writer has viewed the page and read it.

  • Asking a model for statistics, dates, prices or study findings.
  • Having a model summarize a paper, filing or transcript for you.
  • Copying a citation a model generated, even with the intent to verify it later.
  • Using a model to translate a document the author is not able to read independently.
  • Treating a generated search summary as the thing that was actually said.

How the No AI Research Policy Handles Sources

Every draft includes a source list, and the list is not to be ignored. Each claim must be mapped to an entry on the list: the name of the publisher, the title, the date the claim was opened, and the page or timestamp of the claim. A reviewing editor selects a few of the entries, opens the files, and reviews the entry before the draft is submitted to any client.

Primary sources always trump or outrank secondary sources. When a trade magazine publishes a figure from a government report, the journalist checks the report. Magazines that take information from other sources and publish it without citation are more likely to be leading sources than cited sources. Direct quotes are compared word for word to the recording, transcript or published information, as well as the sentences surrounding the quote.

The False Citation is the main risk this policy is designed to cover. A model will generate an author, a journal, a volume and a year that resemble a citation and belong to no paper that has ever been published. The formatting is spot on, which is why it is the problem. Nothing about it looks incorrect, until someone tries to find it.

Ask for the source list

Clients can request the original sources behind a completed project, and we are obligated to provide them. If a source is found to be inaccurate, a correction is warranted; not an argument.

Why the No AI Research Policy Improves Accountability

Being accountable means being able to explain where a statement comes from, and it’s essential when a source sends humans to do the reading instead of machines. Machines cannot be made to explain themselves because the writer is left to reconstruct the reasoning that never happened. The client ends up with only the published page after that.

We will not argue what the policy shows. There’s no test that proves a writer never opened a chat window during research. We have a checkable source list, an editor who checks part of the list for every single order, and a chat window that ends the day fabricated evidence appears in a draft.

This combination alters the incentive more than the technology. A writer who creates their own citation is rarely detected by a determiner score. They are detected when a reader clicks the link. Since the writing rate is $10 per 100 words and 100% of the writing price (what the client pays) reaches the writer, there is a budget to read the source properly.

  • Each argument in the text is endorsed by its writer and editor who are named.
  • The sources are marked with the date on which they were accessed, so you will know when the data in the document was current.
  • Corrections are issued in writing rather than patched in silence.
  • A fabricated source ends the writer’s engagement with Articled.

No AI Research Policy FAQs

Yes, and its expected. Search engines, academic databases, company filings, archives, and paid subscriptions are typical tools. The one exception concerns the generated summary currently placed on top of the search results page. A writer may need to read that summary in order to find that underlying page, however the claim must be sourced from that page.

A reviewing editor will manually check some of the source list. They will not rely on the formatting of the citations. Instead, they will check the wording of the quotes and see where the figures come from in the secondary article.

Not normally. The reports and detector always go out with each delivery email, and the source list stays with the editor and is sent as requested, usually on the same working day. Clients ordering reports, white papers and anything data heavy tend to ask for it, and we appreciate that.

You can point this out to us and we will address it. If we find the underlying source to be incorrect, we will provide a correction along with a revised version, at no charge, during or after your fourteen-day revision window. We will not charge you to revise a round that you already purchased, if we are at fault.

Automated transcription is accepted as a typing aid. It captures spoken words for a record. It doesn’t capture new ideas. Writers listen along and edit the transcript to cite it. Summarizing that transcript with a model is not accepted.

A writer should read original works in their original language. If a writer can’t read a source language, they will either be removed from that segment or paired with a writer that can read that language. Translating a source text with a translation program and quoting that result is using AI-generated text, and is therefore prohibited.

Yes, the primary source references included in the reference section are useful to locate primary sources. It is problematic for a statement in your narrative to be supported solely by a Wikipedia statement. The writer checks the citation and cites the original.

Research is the secret sauce of every project. No one can see it, but everyone recognizes its importance. A piece can still read beautifully, but Green-lighting a piece based on stats can be problematic, and the statistic it rests upon inexplicably traces back to an unsubstantiated source. Preparing our work takes a little longer, but we want to provide you with something that you can verify the source for every stat.

Content a person actually wrote

$10 per 100 words, and the writer keeps all of it. Our 1% sits on top, 0.5% goes to trees, and no generated text appears anywhere in the process.