User Guide: Citizen Analyst for Citizen Journalists
Citizen Analyst Documentation

User Guide: Citizen Analyst for Citizen Journalists

Skip to main content
Table of Contents
< All Topics
Print

User Guide: Citizen Analyst for Citizen Journalists

Last updated: November 15, 2026

Who this guide is for: independent reporters, newsletter writers, freelancers, student journalists, local news staff with no research desk, and community watchdogs who report on government and the people and companies around it.

What it covers: what Citizen Analyst can do for your reporting, how to use the knowledge base and the AI chatbot together, thirteen worked use cases with sample questions, how to handle numbers, data and maps, the public datasets worth knowing and the caveat each carries, and the editorial rules that keep your work defensible.

Not legal advice. Nothing on Citizen Analyst, including the chatbot’s answers, is legal advice. If you face a subpoena, a legal threat, or a source in danger, contact a lawyer. Emergency contacts are at the end of this guide.


1. What Citizen Analyst is, in one paragraph

Citizen Analyst is an independent civic publication and research platform at citizenanalyst.ai. It publishes a knowledge base of more than 1,500 documents: sourced accountability profiles of public officials and companies, investigative methodology, know-your-rights material, voting guides for all 56 US jurisdictions, and backgrounders. It adds an AI chatbot that answers questions from that knowledge base. No account is needed, and the site does not ask for your name or email. What happens to chat messages is set out in How Citizen Analyst Uses AI and the Privacy Policy. The full feature reference is the Citizen Analyst User Manual; this guide is about using those features to report.


2. Who you are: three citizen-journalist personas

These personas are illustrative. Find the one closest to you; the use cases in §5 note which persona each one serves.

Persona A: the local watchdog

Who A part-time newsletter writer or volunteer reporter covering one county, city or school district. Often the only person covering a beat that a closed local paper used to cover.
Goals Explain what local officials and their federal representatives are doing, who funds them, and where public money goes.
Constraints No research budget, no lawyer on retainer, limited time. Needs methods that work with free public records.
Biggest risks Overstating a pattern (a donation is not a bribe), naming private people who should not be named, and legal threats from the people covered.
Starts with Accountability profiles of their members of Congress and state officials; the campaign-finance and FOIA skills; the public-meeting method; the voting guide for their state.

Persona B: the independent investigator

Who A freelancer or Substack investigative reporter working a long project: a contractor, an agency, a network of political operatives. Comfortable with spreadsheets and public databases.
Goals Build a document-backed story that survives legal review and gets picked up by larger outlets or by congressional staff.
Constraints No subpoena power, so everything comes from public records, voluntary sources, and material others have already forced into the open.
Biggest risks Chain-of-custody gaps, unverified leaks, and correlation presented as causation.
Starts with The investigative toolkit, the oversight and investigations skills, the corporate-intelligence and network skills, the data-diary method, and the profile network graphs.

Persona C: the verifier

Who A fact-checker, open-source researcher or community-media volunteer who checks viral claims, videos and official statements.
Goals Say quickly and accurately whether a claim, image or quote holds up, and show the work.
Constraints Speed. Claims spread faster than checks.
Biggest risks Calling a misquote a lie, or calling a lie a misquote.
Starts with The media-verification skills and tutorials, the web-archiving method, the spokesperson and official profiles, and the chatbot for fast orientation.

3. What you can use, and what each piece is for

Resource Where What it gives a reporter
Search citizenanalyst.ai/search Searches the whole knowledge base, highlights matches, and understands acronyms (DOJ, J6, SCOTUS, 1A). Start here when you know what you are looking for.
AI chatbot The chat window on the site Answers plain-language questions from the published knowledge base and may cite the documents it drew on. Start here when you are not sure what exists. See §4.
Accountability profiles Knowledge Base → People and Organizations About 940 people and 65 organizations as of November 2026: officials, legislators, appointees, operatives, media figures, corporate board members and companies. Each profile cites its sources, and profiles of private persons carry a Basis for Inclusion block that states why the subject is included.
Investigative trails Inside many profiles A “where to look next” block listing the public-records systems (court dockets, filings, disclosures) relevant to that subject. See the trails protocol.
Network graphs Inside profiles and on graph pages An interactive map of how a person or organization connects to others in the knowledge base. Connections labelled AI-Generated are leads, not verified claims.
Timelines, infographics, charts Inside articles Dated chronologies and charts, each tied to its sources. Images our editing console generates with an AI image model carry a visible “AI-generated illustration” label. See §9.
Glossary citizenanalyst.ai/glossary Plain-language definitions of legal and civic terms; hover over a highlighted term in any article.
Methodology skills Knowledge Base → Investigative Toolkit → AI Skills 110 published methodology articles. Each is a step-by-step method written so a person can follow it. Recent additions cover checking numbers, keeping a data diary, finding failed PDF redactions, monitoring public meetings, scraping ethically, mapping responsibly, accountability work on policing and immigration enforcement, and testing whether local news reports are independent.
Tutorials and methodologies Knowledge Base → Investigative Toolkit Hands-on tutorials (FOIA, OpenSecrets, PACER, OpenCorporates, reverse image search, web archiving) and tool-agnostic methods. Index: Investigative Research Tools.
Tool directory Investigative Journalism Online Tools Directory Tools by function, with what each costs and what it is good for.
Scoring instruments Methodology pages Documented, published methods for scoring conduct, statements, rhetoric, legislation and agency damage. Read the method before you quote a score. See §6.
Trump Autocracy Tracker citizenanalyst.ai/tracker A 20-marker dashboard of democratic-backsliding indicators, with a public scoring methodology. Findings are queued for editor review and the pages are republished by editors, so check the “Last Updated” line before you cite a score.
Voting guides Voting & Elections Registration, ID, early voting and vote-by-mail rules for all 56 jurisdictions. Useful for election-season explainers.
Corrections log Corrections and Transparency Every substantive correction we make, dated. Check it before you rely on one of our documents.

4. Using the AI chatbot well

The chatbot is the fastest way into the knowledge base. It retrieves the published documents most relevant to your question and answers from them. The embeddable Rights Advisor backend does not save conversations; the floating chat button keeps your recent messages in your own browser. Do not paste source material, names of sources or personal details into either. How Citizen Analyst Uses AI lists which companies receive your messages.

What it is good at

  • Orientation. “What do you have on X?” when you do not know the right search term.
  • Method. “How do I trace who owns this LLC?” returns the steps from the corporate-intelligence method, with the tutorials that go with it.
  • Explanation. “What is a periodic transaction report and when is it due?” or “What can a congressional committee actually compel?”
  • Drafting help. Outlines for a records request, questions for an interview, a structure for an explainer.

What it is not good at

  • It is not a source. Cite the primary document the chatbot points you to, not the chatbot. If it gives you a fact with no document behind it, treat the fact as unverified.
  • It can be wrong or out of date. Open the cited article and check its date and its sources.
  • It mostly answers from what we have published. The floating chat button may also search a vetted list of news and reference sites. Neither is a source: open the cited page. Something that happened this morning may not be in either.
  • It cannot see password-protected articles. Those are excluded from the chatbot by design.

How to ask

Instead of Ask
“Tell me about Senator X” “What does the Citizen Analyst profile of Senator X say about their committee assignments and campaign finance, and what sources does it cite?”
“Is this company corrupt?” “Walk me through the steps to check a company’s federal contracts, lobbying filings and beneficial owners, in order.”
“Write my story” “Give me an outline for an explainer on how a FOIA appeal works, with the documents on your site I should read first.”
“Did the official lie?” “What is the difference between a misquote and a misleading statement, and how does your spokesperson method decide which one this is?”

Privacy reminder. Do not paste a confidential source’s name, unpublished documents, or anyone’s personal data into any chatbot, ours included. Use the chatbot for method and orientation; keep sensitive material in your own secure workspace.


5. Thirteen use cases, step by step

Each use case lists the persona it serves, the steps, the documents to use, and sample chatbot questions.

Use case 1: Who funds my member of Congress, and does it line up with their votes? (Persona A)

  1. Open the profile. Search the member’s name. Federal legislators are under People → Federal Legislators. Read the sourced record and the investigative-trails block.
  2. Pull the money. Follow the Campaign Finance Transparency method for FEC Schedules A, B and E, and the OpenSecrets tutorial for the industry view.
  3. Check the lobbying side. The Lobbying and FARA Disclosure Analyst method shows who is paid to lobby on the bills in front of the member’s committees.
  4. Run the timing test, the skill’s method for comparing contribution dates with votes and hearings. Record every gap in days, both directions.
  5. Write it carefully. The skill’s language rules apply: a donation pattern is not a quid pro quo. Report the timing as timing.

Ask the chatbot:

  • “Explain FEC Schedule A, B and E in plain language and tell me which one shows outside spending for or against a candidate.”
  • “What does the campaign-finance timing test compare, and what words should I avoid when I report the result?”

Use case 2: A company in my town just won a large federal contract (Personas A and B)

  1. Identify the company properly. OpenCorporates and beneficial ownership tracing establish who owns it.
  2. Follow the money. The Corporate Intelligence Investigator method covers registries, sanctions screening, FEC, USAspending and SEC filings. The corporate investigation methodology is the tool-agnostic version. For a nonprofit contractor, its IRS Form 990 filings are searchable in ProPublica’s Nonprofit Explorer (§8).
  3. Treat name matches as leads. A sanctions or watchlist search that returns the company’s name shows that a name matched an entry, not that your company is sanctioned. Confirm identity from the underlying record before you write either word.
  4. Check for an existing profile. Search Organizations; check the network graph for links to officials already in the knowledge base.
  5. Look for influence. Lobbying filings by or for the company, and political contributions by its executives (use cases 1 and 3 methods). Relationship databases such as LittleSis, and domain-ownership lookups such as reverse WHOIS, give leads only: follow each to the SEC, FEC or court record and cite that. A WHOIS registrant is self-declared, often a privacy proxy, and the record can be stale.
  6. Ask for the records. Request the contract file, solicitation and evaluation under FOIA (use case 4).

Ask the chatbot:

  • “What are the steps to trace a company’s beneficial owners using free tools?”
  • “Which public records show what a federal contract was for and how it was competed?”

Use case 3: Verify a viral video or image (Persona C)

  1. Preserve first. Archive the post before it disappears: web archiving for evidence. Capture the page you read, not a later version (see use case 12).
  2. Verify. The video verification checklist, reverse image search and geolocation from photos. The Media Verification Specialist method is the full version, including synthetic-media checks.
  3. Check provenance, not a detector. AI-image detectors are not evidence: in a May 2026 NewsGuard audit, five leading detectors called authentic war photographs AI-generated 13.33% of the time. Check the image for Content Credentials (C2PA) with a free viewer, find the earliest copy with a reverse image search, and read the image itself (shadows, text, reflections). A valid credential shows who signed the file, not that the scene is true; a missing credential proves nothing, because re-saves and many uploads strip it.
  4. Use shadows and maps as leads. Chronolocation tools such as ShadeMap simulate the shadows buildings and terrain cast at a given time and place; use them with a second method. A community map such as GeoConfirmed shows one volunteer view of where something was filmed. Cite the underlying media and your own geolocation, not the pin.
  5. Protect the people in it. Bystanders are private citizens. Blur faces you do not need, and strip metadata before sharing your own footage: metadata stripping.
  6. Add the rights context. The know-your-rights articles explain what the law says about some encounters, and What to Do When You Witness a Constitutional Rights Violation covers filming and witnessing. The knowledge base may not cover your state, so confirm local law with counsel.

Ask the chatbot:

  • “Give me a checklist for verifying where and when a video was filmed.”
  • “What are the legal rights of someone filming police in a public place?”

Use case 4: Run a records campaign on a local or federal agency (Personas A and B)

  1. Plan before you file. The FOIA and Sunshine Laws method tells you to mine existing FOIA logs and prior releases first, route the request to the right component, and build the request the way a congressional committee builds a document-request letter: definitions, custodians, date ranges, preservation, native format.
  2. State records. The same skill has a method for state public-records laws, so you do not need fifty separate guides.
  3. File well. Filing effective FOIA requests and the FOIA filing guide cover fee waivers, news-media status and expedited processing.
  4. Track everything. Log every request, deadline and page received, with your own control numbers. The skill includes a tracking table.
  5. Read the response. Withholdings and redactions are findings in themselves. The skill shows how to test them, appeal, and use OGIS mediation, and when to call a lawyer. Check every PDF you receive for redaction that was never applied (use case 11).

Ask the chatbot:

  • “How do I write a FOIA request that is specific enough not to be rejected as overbroad?”
  • “My FOIA request was denied under Exemption 5. What does that mean and how do I appeal?”

Use case 5: Cover a congressional hearing from home (Personas A and B)

  1. Before the gavel. The Hearing Prep and Analysis method produces a hearing memo (purpose, jurisdiction, witnesses, documents in play) and a list of questions the committee should ask, each tied to a document.
  2. Know the witnesses. Check whether the witness has a profile; a spokesperson’s record is scored with a published method.
  3. During and after. Sort every answer: answered, partial, deflected, “I’ll get back to you,” contradicted by the record, or Fifth Amendment invocation.
  4. Check contradictions properly. Confirm the earlier statement is quoted accurately and in context before you call an answer a contradiction.
  5. Follow up. Track questions for the record (QFRs) and on-the-record promises until they are answered or missed. That follow-up story is often better than the hearing story.

Ask the chatbot:

  • “What can this committee compel, and what happens if a witness refuses?”
  • “What is a question for the record, and how long do agencies usually have to answer?”

Use case 6: A source wants to blow the whistle (Persona B)

  1. Stop and protect. Move to a secure channel: the Secure Source Communication method and the source-protection methodology.
  2. Keep source audio off the cloud. Audio from a confidential source, or any recording that could identify one, is never uploaded to a cloud transcription service or a general-purpose AI chat: the provider then receives the audio, and deletion is not immediate. Transcribe it by hand, or with an open-source speech-to-text model such as Whisper that runs on your own computer with no upload. Citizen Analyst’s editors do the same for source audio. Cloud transcription is acceptable for public-record audio only, such as a hearing or a speech. Check every quote against the original audio before you publish it.
  3. Check what you send out. Before you return or publish any PDF, run the outgoing checks in use case 11.
  4. Know your role. The Whistleblower Protections method separates the journalist’s role from the advocate’s role. You can give a source the neutral referral list and the names of the official channels, and tell them to see counsel before sending you anything. You cannot tell them whether the law protects them.
  5. Route to counsel. The skill maps the federal statutes, protected channels, filing deadlines and remedies, and the congressional channels, naming both majority and minority offices, never one party.
  6. Verify independently. A whistleblower’s account is a lead. Corroborate it against documents and other sources.

Ask the chatbot:

  • “What should I tell a federal employee who wants to report wrongdoing to Congress?”
  • “Which channels are lawful for disclosing classified information?”

Use case 7: Check an official’s public statement (Persona C)

  1. Find the record. Search the official’s profile; many include sourced statement records.
  2. Use the right instrument. A spokesperson’s statements are scored with the Spokesperson Misinformation Assessment; an official’s actions use the Democratic Malice Assessment. They measure different things; see §6.
  3. Cite the claim and the evidence, both to primary sources. The Citation Checker method covers quotation accuracy.

Ask the chatbot:

  • “What is the difference between your spokesperson misinformation score and your malice score?”
  • “What primary sources does the profile of [official] cite for [claim]?”

Use case 8: Report the local effect of federal agency cuts (Persona A)

  1. Start with the tracker. The Agency Damage Assessment Tracker lists scored federal agencies.
  2. Read the method. The Agency Damage Assessment method explains what is scored and from what evidence.
  3. Mind the data. Before you cite federal spending or workforce figures, read the dataset caveats in use case 10 and §7: USAspending’s agency data is top-tier only, and OPM counts people, not full-time equivalents.
  4. Localize. Find the field offices, grants and programs in your area through USAspending and the agency’s own pages.
  5. Check the watchdogs. The Government Oversight Mechanisms method shows how to find relevant GAO, inspector general and CRS reports and the open recommendations agencies have not acted on.

Ask the chatbot:

  • “Which inspector general or GAO reports cover [agency], and what recommendations are still open?”
  • “How does the Vacancies Act affect who is running [agency] right now?”

Use case 9: Cover a city council, county board or election canvass from home (Personas A and B)

  1. Find the record. Many cities publish agendas, votes and legislation through a Legistar site at .legistar.com. Others post PDFs on the city’s own site. Minutes, once approved, are the record of a vote.
  2. Find the video. The LocalView database lists local-government meetings captured on YouTube, with dates and auto-generated captions. It runs about six months behind the present, it cannot tell you when the next meeting is, and its captions and place labels are machine-made. Treat a hit as a lead and confirm it against the video. Quote only a short excerpt and link the video.
  3. Never quote an AI summary without the clip. An AI summary, a machine transcript or a vendor’s highlight is a tip. A quote, a vote or a statement of what a body decided goes into your draft with the video link and the start and end time, checked by a person who watched that clip. Take votes from approved minutes.
  4. Use Documenters notes properly. Notes from Documenters.org are published under a Creative Commons Attribution 4.0 licence: credit Documenters.org, the Documenter and the local partner, with a link. The site’s terms forbid automated copying, so read the notes in a browser and do not scrape them.
  5. Public commenters are private people. A resident who speaks at public comment is on the record but is not a public official. Before you name one beyond the meeting record, and always for a minor, apply the private-citizen standard in rule 3 of §9. Reach them as a source rather than quoting the clip alone.
  6. At each election, watch county certification. Read each county’s certification deadline in the state statute itself, on the legislature’s website. The National Conference of State Legislatures’ summaries are a finder for the citation, not a source for the deadline. Watch board agendas for canvass and certification items, and cite the minutes. A delay is not a crime, and an indictment is not a finding: report the vote, each member’s stated reason from the clip, the deadline, and what the statute or a court says happens next.

Ask the chatbot:

  • “How do I find out whether my county board has put the election canvass on its agenda?”
  • “What is the difference between a county canvass and state certification?”

Use case 10: Report on policing or immigration enforcement (Personas A and B)

  1. Start with the method. The Law Enforcement Accountability Investigation method sets out which records exist, what each one proves, and the limits on naming people.
  2. Gate the names. Before you name any deputy, officer or ICE or CBP agent who is not elected or Senate-confirmed, apply the private-citizen standard in rule 3 of §9: the badge alone is not an anchor. The anchor is documented conduct in an official record, a court finding, congressional testimony or credible journalism. Never identify a masked or unnamed officer or agent by face search, reverse-image search, a people-search site or a data broker. Identity comes from a court filing, an agency release or an official roster.
  3. Use the primary record for 287(g). ICE’s own spreadsheet of participating agencies is the primary record. Download it, note the “as of” date on the page, and cite that date. An agreement shows the legal scope of a partnership, not arrests, detentions or transfers. The Deportation Data Project adds history and geography on top of ICE’s list and says so itself.
  4. Read the ICE datasets as what they are. In the Deportation Data Project’s ICE tables, a row in the arrests table is an arrest, not a person, and the data cover only ICE Enforcement and Removal Operations. A detainer row is a request, not proof a jail complied. The project warns of reliability problems in the removals data across several recent releases, so do not build a removals trend across them without saying so. Cite the dataset and its release date, never the tool you used.
  5. Do not read silence as evidence. Since ICE Directive 11003.7 of 2 June 2026, deaths within 30 days of release are no longer reported. A name missing from ICE’s detainee-death page proves nothing.
  6. Use police-killing datasets as leads. Mapping Police Violence is compiled from media and public records. Confirm any case against the official record, cite the dataset with the date you accessed it, and note that its coding scheme changed in July 2026, so older exports do not line up with newer ones.

Ask the chatbot:

  • “What does an ICE 287(g) agreement allow, and what does the public list not tell me?”
  • “What records show whether a county jail honoured an ICE detainer?”

Use case 11: Work from a PDF you received, or one you are about to publish (Persona B)

  1. Check what you received. A black box over text proves nothing about the text under it. Check every PDF you receive for text under black boxes, redaction marks that were never applied, images under boxes, text hidden under scanned pages, earlier revisions kept by incremental saves, metadata, hidden layers and attachments.
  2. Know the tool’s limit. Free Law Project’s X-Ray finds only boxes drawn over text still in the file, and its own authors say it handles only the most basic cases. A clean result does not prove a document is safe.
  3. Stop before you read further if recovered material is on its face classified, privileged or sealed by a court, or names a victim, a minor or a confidential informant. Never publish a victim’s, minor’s or private person’s name recovered from someone else’s failed redaction. Questions about whether you may publish go to a lawyer.
  4. Check your own file before you publish. Redact by removal, with a tool that deletes the content under the mark, never a drawing tool, a highlighter or a black font. Then run X-Ray and a select-all-and-copy test on the final file.
  5. Keep it local. Do not upload a source document, an unpublished production or a draft to an online redaction checker or PDF service. The service then holds the document.

Ask the chatbot:

  • “How do I check a PDF I received for redactions that were never applied?”
  • “What should I do if I recover a victim’s name from a failed redaction?”

Use case 12: Document a government web page that changed or vanished (Personas A and C)

  1. Capture before you cite. Pages move and disappear, often after your story runs. Capture the page at the time you read it, with the Internet Archive’s Save Page Now or a comparable service, and cite the capture beside the live address. Do not rely on archive.today: Wikipedia blacklisted it in February 2026 over altered archives. It can help you find a page; it should not stand behind a claim.
  2. Use the archive’s index for what it is. The Internet Archive’s capture index lists the dates a page was captured and when its content changed. It shows changes, not visits. You can look a page up and cite it; do not use it to bulk-download captures.
  3. No snapshot is not evidence of absence. The archive holds what crawlers and users happened to capture. “No capture” means “not in the index”, never “the page did not exist” or “the page never said it”. A 403 is not a deletion either: it can be a bot wall, a rate limit or an outage.
  4. Triage the change before you write. A removal can be a redesign, a system migration, routine archiving, legal review, a correction or plain error. Look for the ordinary explanation first, and ask the agency.
  5. Report what the captures show. A page edit is not proof of intent. Attribute motive only to a document or an on-the-record source.
  6. Page-change alerts are a trigger, not evidence. Third-party services watch a page and alert you when it changes; some are free and some are paid, and changedetection.io is an open-source one you can run yourself. Capture the page before you cite it.

Ask the chatbot:

  • “How do I show that a government web page changed, using archived copies?”
  • “What are the ordinary reasons an agency removes a web page?”

Use case 13: Check whether a local story is independently confirmed (Personas A and C)

  1. Count independent reports. Two reports are independent only if they have different owners and different origins. Syndicated stories, stories from the same chain, must-run stories and rewritten stories count once.
  2. Flag the pseudo-local networks. Sites in the Metric Media and Local Government Information Services networks never count as corroboration. A news-index hit, for example from GDELT, shows that an outlet ran a story, not that the story is true; it is a lead.
  3. Read the gap as a gap. Where local coverage is thin, a missing report is not a missing story. Northwestern’s State of Local News project maps where coverage has closed or thinned.
  4. Go to the record holder. Confirm the fact against the court docket, the agency or the minutes, not against another article.

Ask the chatbot:

  • “How do I tell whether two local news reports are really independent?”
  • “What is Metric Media, and why can’t I count it as corroboration?”

6. Reading our scores before you quote them

Citizen Analyst publishes six scoring instruments. Each measures a different thing. Quoting one as if it measured another is the most common error we see.

Instrument Measures Unit Method
DMA (Democratic Malice Assessment) A person’s actions One person DMA methodology
SMA (Spokesperson Misinformation Assessment) A spokesperson’s statement accuracy during their tenure One spokesperson SMA methodology
DRA (Dangerous Rhetoric Assessment) A public figure’s public speech One public figure the Dangerous Rhetoric Assessment skill
LIA (Legislative Impact Scoring) One bill, executive order or rule, on two tracks One action the Legislative Impact Scoring skill
WLM (Winners and Losers Mapping) Who gains and loses in a market across many actions One market the Policy Winners and Losers Mapping skill
ADA (Agency Damage Assessment) Harm to one federal institution One agency the Agency Damage Assessment skill

How to quote a score: name the instrument, link the method, and say what it measures. For example: “Citizen Analyst’s Democratic Malice Assessment, which scores documented actions, not statements, rates…”. The full inventory is Scoring and Inclusion Systems.


7. Numbers, data and maps

Numbers

Before you publish a number, run the Numbers and Statistics Integrity checks on it.

  • Name the base and the year. Say what a percentage is a percentage of (“42% of the 1,200 detainees”), and give a rate its denominator and year (“per 100,000 residents, 2024”).
  • Use points for differences between percentages. A rise from 10% to 12% is two percentage points (a 20% relative rise).
  • Give both figures when you say “X times”. A ratio of two small counts reads as large.
  • Give absolute risk with relative risk.
  • Adjust dollars across years for inflation, or say that you did not.
  • Do not call a lead or a trend that sits inside the margin of error.

Keep a data diary

Start a data diary with the first download: the source and release, its flaws, every transform you apply, and the query or script that produced each figure. Hash raw files and never edit them in place. Before you publish, have a second person re-run the analysis from the diary alone. Then read the dataset’s caveat card against your wording. The Data Diary and Reproducibility method carries cards for three datasets reporters cite often:

  • Deportation Data Project (ICE). Rows in the arrests table are arrests, not people, and the data cover only ICE Enforcement and Removal Operations (see use case 10).
  • USAspending. The agency-level data covers top-tier agencies only, and account data does not exist before fiscal year 2017.
  • OPM federal workforce data. It counts people, not full-time equivalents, and excludes bodies OPM does not cover, such as the Postal Service and the intelligence agencies; “not covered” is not zero. The system changed in January 2026, so a series that joins old and new figures crosses a break; say so.

Maps

The Ethical Mapping method applies to every map, and it should apply to yours.

  • Map rates, not counts, when places differ in population.
  • Never map where vulnerable people live or shelter. For data about vulnerable people, aggregate to the county or larger, and suppress any area whose count is 1 to 9.
  • Disclose the class breaks, use colour-blind-safe colours, and cite the boundary file and its year.

8. Public datasets worth knowing, and how to treat each

These are third-party sources. Some are tools Citizen Analyst’s editors use; the download, the analysis and the caveats are yours to handle. The right-hand column is the one to read first.

Source Good for Treat it as
ProPublica Nonprofit Explorer IRS Form 990 filings: revenue, expenses, officer pay A finding when you cite the filing. Returns lag the tax year by a year or more.
OpenSanctions Sanctions, politically-exposed-person and watchlist matches A name match, unverified: confirm identity, then cite the underlying listing.
GDELT Finding which outlets ran a story A lead only. It indexes what outlets published, machine-tagged.
LittleSis Who sits on boards and funds whom A lead only. Relationships are crowd-sourced; follow the source it cites and cite that. Its data is tagged CC BY-SA 4.0, so check the terms before reusing a table.
Whoxy and reverse WHOIS Links between websites through a shared registrant A lead only, and a paid service. Registrants are self-declared.
Open Payrolls Public-employee pay records Useful for officials and for patterns across an agency. Pay of a non-official public employee is not a story by itself, and a private person needs the standard in §9, rule 3, before any profile.
ICE 287(g) participating-agencies list Which agencies hold an agreement The primary record. Download the spreadsheet by hand and keep its as-of date. It shows scope, not arrests.
Deportation Data Project ICE arrests, detainers, detentions, removals, facilities and 287(g) history A secondary compilation of ICE data, CC0 licensed. Download by hand and cite the release date. See use case 10.
Mapping Police Violence People killed by police A lead to the official record. Download the CSV by hand; cite it with the date accessed.
EAVS (Election Administration and Voting Survey) How jurisdictions run elections: rejected and provisional ballots, poll workers A government dataset that each jurisdiction reports for itself. Download by hand and read the codebook. Missing values are not zeros: a blank, or a coded non-answer, means no figure.
Legistar A city’s agendas, votes and legislation What the clerk already publishes. A matter’s title is the clerk’s text, not a summary of what a law does.
LocalView Local meeting video and auto-captions A lead. About six months behind; see use case 9.
Documenters.org notes Plain-language notes on local meetings CC BY 4.0: credit them. Read in a browser.
OpenStreetMap (Overpass) Mapped features such as buildings and roads A geolocation lead: a missing feature proves nothing and a present one needs confirming in imagery. Credit OpenStreetMap and note the ODbL licence.
Chronicling America (Library of Congress) Historic US newspapers to 1963 Find the page with it, but the OCR text is often wrong: quote from the page image. Check copyright: the collection runs to 1963 and the Library does not own the rights.
Internet Archive Wayback Machine Captured copies of web pages Look-up and cite only. No snapshot is not evidence of absence (use case 12).
Who Posted What Keyword searches across social platforms A lead. Results are each platform’s own search, and filters apply loosely; capture what you cite.
Free Law Project X-Ray Detecting failed redactions One test among several (use case 11).
State of Local News (Northwestern Medill) Where local news has closed or thinned Context, view-only.

We have not built in Vera’s Incarceration Trends, the National Police Index, Arctic Shift, the Police Scorecard API, Good Jobs First’s Violation Tracker, the Corporate Prosecution Registry, OpenSky Network flight data or urlscan.io: their licences or terms bar our use or are not stated.


9. Editorial rules that keep your story defensible

These rules are built into our methods. They will protect your work too.

  1. Allegation, finding and charge are different words. A congressional referral is not a charge. A contempt vote is not a conviction. Trade timing is not insider trading. A donation pattern is not a quid pro quo.
  2. Correlation is not causation. If an agency acted after your story, say it acted after, not because, unless you have documents that show why.
  3. Private citizens get extra protection. Officials, candidates and public companies are fair subjects. A private person needs a documented public role beyond their speech before you profile them. Our standard is the Accountability Profiles Standards FAQ.
  4. A people-search result is a lead, never a fact. Looking up anyone who is not a public official starts with the private-citizen standard in rule 3. Results from people-search sites and data brokers are never cited and never published, and no home address is published. Confirm a lead from a primary record (a court filing, a disclosure, a government register) and cite that.
  5. Police scanner audio is for human listening. Treat it as a lead, never a source of record. Never publish names, plates or addresses heard on air, never identify victims, minors or suspects from radio, and never reveal the location or timing of a live enforcement operation. Silence on a channel proves nothing, because many channels are encrypted.
  6. Cite the primary source. Cite the court filing, the disclosure form or the transcript. Cite Citizen Analyst when you are relying on our analysis, scoring or compilation.
  7. Preserve and hash your evidence. Keep original page numbers, archive web pages when you read them, and never alter a document. See evidence preservation.
  8. Test both sides. When you check one party’s members, check the whole committee.
  9. A pattern found by a tool is a question, not a finding. AI tools and scripts can surface what looks unusual in a dataset. Each such lead is a starting point: confirm it against the record holder before you write it.
  10. Collect data the way a careful guest would. If you scrape, ask for the data first. Read the site’s terms, and be wary of data whose licence is non-commercial or research-only. Honour robots.txt, identify yourself, and go slowly. Stop at a login, a bot challenge or a cease-and-desist. Keep only the fields your story needs, and archive what you fetched. Citizen Analyst’s own collection code is held to these rules.
  11. Label AI images, and never invent a scene. Every image our editing console generates with an AI image model carries a visible “AI-generated illustration” label. Images made since 14 November 2026 on our main editing computer also carry a self-signed C2PA Content Credential, so a verifier shows an unrecognized signer. Earlier images and those from our second editing computer are unsigned, and a crop, resize or re-post strips the credential, so the label is the disclosure that survives. Never use an AI image to depict a real, identifiable person in an invented scene.
  12. Publish your corrections. Ours are at Corrections and Transparency. Tell us when we are wrong: use the contact form.

How we apply these rules ourselves: Rules and Editorial Standards and Editorial Review Process.


10. A suggested first week

Day Do this
1 Read this guide and skim the Journalist learning path. Ask the chatbot three questions about your beat.
2 Find the profiles of the officials you cover. Read one end to end, including its sources and investigative-trails block.
3 Read the FOIA and Sunshine Laws method and draft one targeted request.
4 Work through two tutorials that fit your beat (verification, OpenSecrets, PACER or OpenCorporates).
5 Set up secure communication: Digital Security for Citizens and Keeping Communication Secure. Start a data diary for your first story.

11. Emergency contacts for journalists

Situation Contact
Subpoena for your sources or notes Reporters Committee for Freedom of the Press legal hotline: 1-800-336-4243 (rcfp.org)
Devices seized or digital-security emergency Electronic Frontier Foundation (eff.org)
Press-freedom emergency Committee to Protect Journalists (cpj.org); PEN America (pen.org)
Arrested while reporting National Lawyers Guild (nlg.org): local chapter legal-support hotlines

Related

This guide was drafted with AI assistance. It is for educational purposes and is not legal advice.

Was this article helpful?
0 out of 5 stars
5 Stars 0%
4 Stars 0%
3 Stars 0%
2 Stars 0%
1 Stars 0%
5
Please Share Your Feedback
How Can We Improve This Article?