Data Activists AdvocacyLab

Frequently asked questions

Using the tools

What does CiteCheck do?
You upload a finished report. It lists each quotation with the source the report cites, fetches each source, and tells you whether the quotation is there. Where it is, you see the passage. The guide explains every result.
What does CiteCollect do?
It is for a team writing a report. Each researcher adds quotes as they find them, and each quote is checked against its source right away. A coordinator puts them in order and gets a Word file. Anyone can start one in their own workspace, and in a team its coordinators do; for the Word file to come out in your organisation's own format, get in touch and we'll set up your template.
What does CiteFind do?
You type a research question. It searches for public sources, reads them, and shows passages about your question in each source's own words, with the address. A model chooses the passages, but every word shown is cut from the source. It is not a complete search, so it is a place to start reading.
Does an AI decide whether a quote is real?
No. Ordinary code compares the words: exactly, then ignoring punctuation, then allowing small differences. The row says which of these matched. A model can give a second opinion on a row, such as whether the source says the same thing in other words. It appears as an opinion and can't change the result.
What does the model do, then?
It reads the report and points at where the quotations are and which link each one cites. The code then cuts the words from your file, and only accepts links that are in your file. A model can't write a quote or invent an address, so a row with no address stays empty and asks you for one.
Why not paste the report into a chatbot?
A chatbot's answer reads just as confidently when it never opened the source, and it can differ each time you ask. Any passage it shows was typed by the model, so that can be wrong too. Here each row shows the source's own words, or says why the source couldn't be read.
Can the results be wrong?
Yes. A page may have changed since the report was written, or the link may lead to a page about the document and not the document itself. A close match can also hide a changed word, which is why rows show the words that differ. Treat "Not found" as a reason to open the link, and read the source before you change a citation.
What if a site blocks the check?
The row says so, and this never counts as "Not found", because it tells you about our access and nothing about the quote. Where it can, the tool looks for a dated copy in the Internet Archive and shows the date.
What kinds of report does it work on?
PDF and Word files up to 25 MB with readable text. A scan with no text layer is refused. It needs the links to be in the file. We have measured it on country of origin information reports from NGOs. Other reports that quote and link to their sources should work, but we haven't measured them.
What about quotes in translation?
The word comparison doesn't translate, so a translated quote first comes back "Not found". A model can then say that the passage is a translation, and the row shows the source's own words for you to compare.
How long does it take?
It depends mostly on how many sources the report cites. You can leave the page and the work carries on. We'll send you an email when it ends if you've left, unless you turn that off.

Your reports

Who can see my report?
Your CiteCheck tables are yours alone on the site. A CiteCollect report is visible to the people added to it. The few people who run the service can reach the database.
Does my report go to an AI company?
For CiteCheck, yes: the report's text goes to DeepSeek V4.1 Flash (Providers with Zero-Data Retention contracts on OpenRouter), and the upload page says so before you upload. The providers we work with do not store your data or use it for training their models. For CiteCollect, only a quote and passages from its public source can reach a model, and a team workspace can turn that off. For ReportReview, yes: the whole text goes to the models named on its page, and only if you upload it there.
How long do you keep it?
After 30 days the file and its text are deleted, and your table of checked quotes stays until you delete it. You can delete a report and its table at any time. The privacy page has the details, including backups.

Accounts and access

How do I get an account?
Sign up with your email address and a password. We'll send you a link that confirms the address and signs you in.
How do I delete my account?
On your profile page. Your reports and tables go with it. If you are the only coordinator of a report colleagues are working on, you'll be asked to make someone else coordinator first.

Billing

What does it cost?
See pricing. You can start for free, without a card.
What happens when I reach my limit?
New uploads are refused and the page says why. Your existing tables stay as they are, and nothing is charged without you choosing a plan.
Can my organisation pay by invoice?
Yes. Write to advocacylab@dataactivists.org and we'll set it up.