The only service whose output is a verdict

Fact-checking services services
for every claim, a verdict — verified, corrected, or unverifiable.

Fact-checking is the only service here whose output is a verdict. Every claim in the text returns as one of three things: verified against a source at the top of the hierarchy, corrected with the change traceable, or unverifiable — which is itself a finding. The discipline underneath is that the burden of proof sits with the claimer, not the checker, and the failure is never one missed error. It is the quiet substitution of a plausible source for the true one.

Every claim extracted and decomposed before a source is opened. The source hierarchy applied rather than the first result. Three verdicts, including the honest third. Every change traceable to its evidence, named and dated. And a claim ledger that lets the text be checked again after the check.

WHAT THE WORK IS 01 / 04

Every claim returns as a verdict. Verified, corrected, or unverifiable — and the third is the one that saves publications.

  • 01 The claimsExtracted, decomposed.
  • 02 The verdictsThree, one honest.
  • 03 The ledgerTraceable and dated.
The burden sits with the claimer Primary over secondary Three verdicts Every change traceable

The Writely difference

Fact-checking decomposes the claim, applies the source hierarchy, and returns a verdict — never an opinion.

That makes it the only adversarial-to-itself service here. The checker's job is to find what is wrong before the reader does, and the standard is not whether the text reads right but whether every claim survives the source it should have been checked against.

Decompose the claim

'Lagos is the largest city in Africa' is really four questions wearing one coat: largest by which measure, among which class of places, counted how, and as of when. The check that treats it as a single claim checks the part that is easy and ships the rest, which is how failures happen inside sentences that read as one statement. The first pass therefore extracts the claims and breaks them down, and the breakdown is the record of what will be checked — a claim that has been decomposed cannot be quietly half-checked, because the seams are written down. You can also explore academic proofreading.

The source hierarchy, not the first result

Sources are ranked rather than counted. A primary source — the court record, the dataset, the original document, the publication itself — outranks every secondary source, and a source that is merely convenient is not a source at all. Two secondary sources agreeing do not outrank one primary record, and a source that has been wrong before is not settled by confidence. The hierarchy is set at the brief rather than improvised per claim, because the tendency under deadline is always to accept the third result and hope.

Three verdicts, including the honest third

Verified: the record supports the claim, and the source is named. Corrected: the record says something different, and the change is traceable to its evidence. Unverifiable: no reliable record settles it — which is a finding, not a failure, and the options then belong to the author. What does not happen is the quiet downgrade of the third into the first because the checker could not find otherwise. Absence of evidence is not evidence, and the safest-looking error in this whole craft is a claim marked verified because nobody disproved it.

The ledger is the deliverable

A fact check that returns only corrected text has made the publication safe once and stupid forever: nothing survives the check that can be re-checked. So the claim register records what was asserted, what source decided it, when it was checked and where the evidence lives — dated, named, with archived copies of everything that moves. Verification decays, and the ledger is what lets an editor verify the check in five minutes instead of a week. It is also what defends the piece, in public, when a reader challenges it.

Find what your check is missing

Eight things a fact check has to settle, and the version with a search settles three.

None of these are about being fast. They are about whether every claim in the text survives a reader who checks — and whether the check can be repeated, by somebody else, a year later.

The verdict is the deliverable

The output of a fact check is not an opinion and not a confidence: it is one of three verdicts per claim, each with its source and its evidence attached. A checker who cannot return a verdict cannot return a result, and a check that returns 'looks fine' has not been done. The discipline that keeps verdicts honest is the register — every claim written down and decomposed before the sources open, so the check works through a list instead of a memory, and the list is delivered.

Decompose the claim

Fact-checking fails at the seams, inside the sentence that looks like one claim and carries four. The date inside the claim, the measure, the class of thing, the attribution to a source that is itself wrong: all of it is hidden in a statement that reads as a unit, and the checker who treats it as a unit checks the part that surfaces. The decomposition is the moment the check becomes rigorous, and it happens in the register, before a source is opened, on every claim — no exceptions.

The source hierarchy

The ranking is decided at the brief and applied per claim: the primary record over the secondary account, the underlying dataset over the article reporting it, the original text over the quotation of it, the figure with its methodology over the headline number. Where a claim is carried by secondary sources alone, the verdict says that, because the reader who checks will know, and the piece's confidence should not exceed its evidence. Sources that disagree are resolved up the hierarchy, or the claim comes back contested with both sides recorded.

The honest third verdict

Unverifiable is a result, and the one that saves publications. The claim that no reliable record settles comes back named as such, and the author chooses: the claim goes, it is attributed to whoever made it, or it stays with the gap stated in the text. What never happens is a quiet downgrade to verified, because the checker could not disprove the claim and the deadline arrived. The publication that publishes an unverifiable claim as verified has been misled in the safest-looking way, and the reader who checks will do so in public.

The ledger is dated and re-checkable

Verification decays: the figure gets revised, the page moves, the source issues a correction, the title is posthumous. The ledger is written so that the fifth person can re-check the fourth person's check in an afternoon — source named, access date recorded, archive copy kept. What decays quickest gets a re-check date in the plan, and the plan travels with the text. A check that cannot be checked is a promise, and the whole discipline exists so that no promise is needed.

The small facts, where failures live

Fact-checking is not usually about the scandalous claim; it is about the quiet ones nobody opened: the date inside a parenthesis, the spelling of the name, the job title, the population figure, the "well-known" fact. Those are the ones that pass every review because they read as furniture, and they are where the reader's finger lands. The small facts are checked against primary records like everything else, and the register lists them with their verdicts like everything else, because the failure that embarrasses a publication is never the big claim that was debated all week.

Checking is not research

Research finds out what is true when nobody in the room knows; checking tests whether an assertion already in the text is true. The direction of the work is different, and so is the deliverable — a finding produced from nothing, against a verdict returned on a claim. A manuscript that needs its claims found rather than checked needs research first, and a check that quietly does the research has produced a text rather than a verdict. Which of the two a piece needs is decided at the brief.

What this service cannot do

Four limits. Facts are checked, not improved — the correct name and figure go in with their sources, and the argument the author built around them is not touched. Opinions, recommendations and forecasts are not checked; they are classified out of the register, and any mistaken classification is open to dispute. The check cannot promise every claim will settle, because some claims do not settle — that is what the third verdict is for. And it cannot decide whether the piece runs: the ledger goes to the editor, and the editor decides.

Read the claim register against the sources before accepting the corrections. That single pass finds more faults than any review, because a verdict you can trace is a correction you can defend — and a correction you cannot trace is an argument.

A better brief, a better check

Send the text and the deadline, not a wish for accuracy. Everything in the check is decided by what is claimed and what the sources can hold.

The format question comes before the message here too, and in this form the question is what the piece actually claims and where those claims came from — because the claims carried over from the brief are the ones nobody checks. Given the text, the sources and the dates, the verdicts follow without further invention.

  • The text, as it stands.
  • Where the claims originally came from.
  • Any sources the author already holds.
  • What the piece will be used for.
  • The deadline, and the publication date.
  • Which claims are load-bearing.
Share your brief
CHECK LIST Before the verification

Six things before the verification.

Claims extracted
Sources mapped
Hierarchy set
Deadline set
Ledger planned
Re-check planned

How the work is done

From a text to a dated ledger of verdicts the editor can defend.

Extraction, hierarchy, verdicts, ledger. The order is not a preference. A check run without the register is a memory working through a document.

01 / THE EXTRACTION

Pull every claim apart and list it

First, and no source is opened before it. The text is read for what it asserts — dates, numbers, names, titles, quotations, locations, causal claims — and each is broken into its component questions and written into the register with a number. What is opinion or forecast is classified out and listed on their own, so the author can dispute the line where it fell. The claims that arrive from the brief get the same treatment as the ones in the prose, because they are the ones that carry a source nobody opened.

02 / THE HIERARCHY

Rank the sources before opening them

Then the ranking: what the primary record is for each claim, what secondary sources can back up but not settle, and which sources are out of the running entirely — the convenient article, the one that has been wrong before, the statistic without its methodology. The ranking is written down per claim, and the check works from the top of it. Where the hierarchy is thin — the claim lives only in secondary accounts — the verdict records that, because the piece's confidence should not outrun its evidence.

03 / THE VERDICTS

Return three verdicts, and mean them

Then the check itself, claim by claim against the register. Verified is returned with the source and what it actually says. Corrected is returned with the new form of the claim and the evidence that settles it, tracked into the text. Unverifiable is returned as unverifiable, with what was tried and why nothing settled, and the three options that belong to the author. Nothing is quietly resolved — a checker who cannot return the honest third is not a checker, and the publication finds out on the worst day.

04 / THE LEDGER

Deliver the register, the sources and the re-check plan

Finally, the set the editor keeps: the claim register with its verdicts, the source file with citations, access dates and archived copies of every link that moved, and the re-check plan naming what decays and when. The corrected text travels with it, tracked so the author can see each change against its verdict. The publication decision stays with the editor — the desk has checked, and the checking is what was ordered. You can also explore research proposal writing.

What actually gets handed over

Five documents. The claim register is the one that decides whether a fact check exists at all, because a check without a register is a memory working through a document.

The corrected text is the visible part and it comes third. These five are what make the verdicts repeatable — and what defend the text, in public, when a reader checks.

The claim register

Every claim extracted, decomposed and numbered, with the component questions each claim hides written beside it — the date, the measure, the class of thing, the attribution. Opinions and forecasts are classified out and listed on their own, so the author can dispute the line where it fell. No source is opened before the register exists, which is what keeps the check from becoming a memory running through a document and shipping the parts that did not surface.

The verification report

The verdict per claim, with its source, the evidence it settles on, and what the source actually says — the line that was checked rather than the page it appeared on. Where the record says something different from the claim, the differing wording is quoted. Where nothing settles it, the report lists what was tried and why nothing settled, because an honest third verdict is a result that has to be as defensible as a verified one.

The corrected text

The text with every change tracked, each one traceable to the verdict that produced it — the corrected figure beside its basis, the corrected name beside its source, the quoted line beside the passage it came from. The author can see what changed and why, and can dispute a correction by disputing its evidence, which is how the check stays honest. The argument the claims sit inside is not touched, because the argument is not the checking desk's.

The source file

The citations, the access dates, and the archived copies of everything that moved — because sources decay and the ledger that cannot be re-checked is a promise rather than a record. Where a source is secondary only, the file says so, so the ledger's confidence never quietly outruns its evidence. This is the file the editor reads when a reader challenges the piece months later, and it is the difference between a defensible answer and a scramble.

The re-check plan

What decays and when to look again: the figures that get revised, the pages that move, the titles that change, the sources that issue corrections. Each decaying claim carries a re-check date, and the plan travels with the text so that the fifth person can re-check the fourth person's check in an afternoon rather than a week. Verification is not permanent, and the plan is the honesty that admits it.

The claim register is the part senders skip and the only document here that makes the check repeatable. A text corrected without it has been made safe once and left unverifiable by everybody else — which is how a publication gets caught respecting the correction it once made.

What a fact check has to settle

Six things, and the version with a search settles three.

Each can be checked before the publication, and each decides whether the text survives the reader who checks it in public.

  • Every claim extracted, none assumed.
  • The source hierarchy applied, not the first result.
  • Three verdicts, including the honest third.
  • Every correction traceable to its evidence.
  • The ledger dated and re-checkable.
  • What could not be verified, named.
THE VERDICTS Six things

Six things a fact check must settle.

Claims extracted
Hierarchy applied
Verdicts returned
Corrections traceable
Ledger dated
Gaps named

Explore related academic support

The research and the craft that returns a verdict on every claim.

A fact check is a promise that every claim survived a source at the top of the hierarchy. These are the pages that produce and check material of that kind, and every verdict should be traceable to the source that decided it.

Have a draft and a deadline already? Send the text and we will tell you what the check can honestly cover. That is usually fewer claims than the draft carries, and more corrections than it expects.

Where the specifications come from

The Nigerian fact-checking organisation leads, because the practice this page describes runs locally through exactly these desks — and the code that defines it is public.

Two candidates could not be listed: the continental fact-checking organisation and the journalism institute both returned refusals rather than pages, and the rule is that nothing is listed until it answers. The Nigerian fact-checking organisation comes first for a practical reason: the practice of returning verdicts on claims runs locally through desks like this one, and its published work is the standard a check in this market is read against. Then the code of the profession itself — non-partisanship, transparency of sources, corrections — which is the closest thing to a written definition of the job. Then the archive of the discipline and the corrections practice, where the corrections themselves are published with the reasons attached. Then the site that has documented more individual verdicts than any other. Then the two sources that keep numbers and links honest: the statistics house whose datasets any reader can take apart, and the archive that keeps a checked link checkable after it moves. And the plain writing rules, because a verdict that cannot be read plainly cannot be acted on. Four of these eight have not appeared anywhere in this corpus before. Every resource here was confirmed to answer before it was listed. These external resources are informative references, not Writely service pages.

THE LOCAL PRACTICE

Dubawa

The Nigerian fact-checking organisation, and where the practice of returning verdicts on claims actually runs in this market — with its published checks, its corrections and its source lists. Included because a reader who checks a Nigerian publication's claims will measure the work against desks like this one, and because the local record of what has already been settled is the first hierarchy any claim register has to respect.

Open resource
THE CODE OF THE PROFESSION

IFCN Code of Principles

The international fact-checkers' code, and the closest thing to a written definition of the job: non-partisanship, transparency of sources and methodology, and a standing commitment to corrections. Included because this page is a craft page, and the craft has an ethical boundary that a checker crosses the moment a verdict is adjusted for preference rather than evidence. The code is what a checking desk commits to publicly. New to this corpus.

Open resource
THE ARCHIVE OF THE DISCIPLINE

Reporters' Lab

The archive of fact-checking worldwide, and the record of the discipline's methods — which organisations check which subjects, with what transparency, corrected how often. Included because the register's verdicts are better for knowing what has already been checked elsewhere, and because the history of corrections in this field is the clearest evidence of where checks fail. New to this corpus.

Open resource
THE CORRECTIONS PRACTICE

Full Fact

The independent fact-checker, and the public model of what a correction looks like when it is done properly: the original claim quoted, the evidence named, the new wording explained, and the log kept. Included because a correction is a deliverable, not an apology — and the register is the form that keeps it traceable. Also the working demonstration of how a verdict on a claim can be checked by a reader without phoning the publication. New to this corpus.

Open resource
THE DOCUMENTED VERDICTS

Snopes

The oldest library of individual verdicts, and the standing demonstration of the form: one claim, one rating, the sources named and dated, the reasoning visible. Included because the long tail of checking is the same claim resolved a hundred times by a hundred deadlines — and the archive is what lets a desk return a verdict without redoing the work. New to this corpus.

Open resource
THE NUMBERS, TAKEN APART

Our World in Data

The statistics house whose figures any reader can take apart — every chart carrying its source, its method and its units. Included because the most consequential claims in serious writing are numerical, and the difference between a checked number and a copied one is exactly the dataset. A ledger that records the source has to name the one the figure came out of.

Open resource
THE CHECKED LINK, KEPT

Internet Archive

The archive that keeps a checked link checkable after it moves: the archived copy is what turns a verification into a durable verdict. Included because verification decays and the ledger that cannot be re-checked is a promise — and because the source file that records what the page said on the day it was read is the difference between a defensible answer and a scramble.

Open resource
THE ACTUAL RULES

Plain language guide series

Plain rules for plain writing, and the part of this list most likely to be read and used, because a verdict that cannot be read plainly cannot be acted on. The verdict reads as one clause — what was claimed, what the record says, which source decided — and these rules on naming rather than pointing, one meaning per word, and cutting what carries no weight are the pass the register's entries need.

Open resource

Local context, wider reach

Nigeria at the centre. Africa in view.

Writely is rooted in Nigeria and designed for remote academic and business support. Whether you are checking an article, a report, a manuscript, a speech, a claim in a proposal, a data point, a statistic or anything published in the press, from any city or town, or from another part of Africa or another country entirely, you can work on it here.

Remote-friendly writing support

Fact-checking across Nigeria

Explore fact-checking support for writers, newsrooms, publishers and organisations across Nigeria and wherever the text is read. Three things are genuinely different here, and all three belong in the brief before the extraction rather than in a later edit. The claim often arrives third-hand — a figure copied from a blog that copied it from a broadcast — so the first task is finding where a claim actually began, and a claim that cannot be traced to a beginning is a claim that cannot be verified. The reading public is fast and vocal, and a correction that arrives two days late has already been screenshotted by the time it does, so the ledger is written for the reader who checks in public as well as the editor who ordered the check. And the local sources thin out the further a claim gets from the major cities, so the source hierarchy has to be planned rather than improvised: where the primary record does not exist, the honest verdict is unverifiable, and a published guess is worse than an admitted gap.

Lagos Abuja Port Harcourt Kano Ibadan Enugu Benin City Uyo Kaduna Calabar Jos Abeokuta Awka Akure Warri Owerri Zaria Sokoto Bauchi Maiduguri

Nigeria states and the Federal Capital Territory

Writely’s academic support is available remotely across Nigeria, including:

  • Abia
  • Adamawa
  • Akwa Ibom
  • Anambra
  • Bauchi
  • Bayelsa
  • Benue
  • Borno
  • Cross River
  • Delta
  • Ebonyi
  • Edo
  • Ekiti
  • Enugu
  • Gombe
  • Imo
  • Jigawa
  • Kaduna
  • Kano
  • Katsina
  • Kebbi
  • Kogi
  • Kwara
  • Lagos
  • Nasarawa
  • Niger
  • Ogun
  • Ondo
  • Osun
  • Oyo
  • Plateau
  • Rivers
  • Sokoto
  • Taraba
  • Yobe
  • Zamfara
  • Federal Capital Territory

Remote support across Africa

Country-level discovery

Writely welcomes conversations with students, researchers and candidates across Africa, including learners in:

Nigeria Algeria Angola Benin Botswana Burkina Faso Burundi Cabo Verde Cameroon Central African Republic Chad Comoros Côte d’Ivory Democratic Republic of the Congo Djibouti Egypt Equatorial Guinea Eritrea Eswatini Ethiopia Gabon Gambia Ghana Guinea Guinea-Bissau Kenya Lesotho Liberia Libya Madagascar Malawi Mali Mauritania Mauritius Morocco Mozambique Namibia Niger Rwanda São Tomé and Príncipe Senegal Seychelles Sierra Leone Somalia South Africa South Sudan Sudan Tanzania Togo Tunisia Uganda Western Sahara Zambia Zimbabwe

Questions, answered

Start with what you need to move forward.

If your question is not covered here, send the essentials in a chat. A precise brief usually leads to a more useful starting point.

Ask your question
What is fact-checking, and what is it not?
It is returning a verdict on every claim a text makes: verified against a source at the top of the hierarchy, corrected with the change traceable, or unverifiable, which is itself a finding. Not research — research finds out what is true when nobody knows; checking tests claims that are already asserted. Not proofreading, which looks at the language and assumes the content. And not an opinion service: the verdict belongs to the evidence, not to the checker.
What kinds of claims do you check?
The ones that assert a state of the world and can be checked against one: dates, numbers, names, titles, events, quotations, definitions, statistics, locations and causal claims. What is not checked is what the text believes, recommends or argues — that belongs to the author. The first task of the check is deciding which is which, and the decision is written into the register so the author can dispute anything that was classified wrongly.
How do you choose sources?
By hierarchy rather than by availability. A primary source — the court record, the dataset, the original document, the publication itself — outranks every secondary source, and a source that is merely convenient is not a source at all. For a number, the underlying dataset outranks the article that reported it. For a quotation, the original transcript outranks the news story. Where only secondary sources exist, the verdict records that, because a claim carried by a secondary source alone is weaker than it looks.
What happens when a claim cannot be verified?
It comes back as unverifiable, which is the third verdict and the one that saves publications. The options then belong to the author: the claim goes, it is attributed to whoever made it, or it stays with the gap named in the text. What does not happen is a quiet downgrade to verified because the checker could not find otherwise — absence of evidence is not evidence, and a checker who resolves the third verdict into the first has misled the publication in the safest-looking way there is.
How do you handle sources that disagree?
By going up the hierarchy rather than counting heads. Two secondary sources agreeing do not outrank one primary record, and a source that has been wrong before is not settled by confidence. Where the disagreement survives the hierarchy — two primaries genuinely conflict — the claim comes back as contested, with both sides recorded, and the text either states the disagreement or drops the claim. A check that hides a live disagreement has picked a side and called it a verdict.
What is the difference between fact-checking and research?
Research answers a question nobody has answered in the document; checking answers whether an assertion already in the document is true. The direction of the work is different: research starts from nothing and produces a finding, while checking starts from a claim and produces a verdict against it. The skills overlap — the sources, the scepticism, the records — but the deliverable does not, and a piece that needs its claims found rather than checked needs research first.
What is the difference between fact-checking and proofreading?
Proofreading assumes the content is true and fixes the language; fact-checking assumes the language is fine and interrogates the content. The passes are complementary and the order matters, because a corrected sentence that is then rewritten is a wasted correction. What proofreading catches is the grammar; what checking catches is the fact that reads perfectly and is wrong — the most dangerous fault on any page.
How long does fact-checking take?
As long as the number of claims times the depth of the hierarchy, which is longer than writers expect, because decomposing the claims is a reading task and the sources do not always answer instantly. The honest schedule is set after the extraction, when the claims are counted, and it is governed by the publication date. What the work will not do is promise a same-day check on a document carrying a hundred claims, because the fast version is the one that checks the easy claims.
Do you check opinions and predictions?
No. An opinion is not a claim about the world; a prediction is not yet checkable; a quotation of what somebody said is checked for accuracy of quotation, not for truth of content. What the check does instead is split the three into their kinds: the factual assertions inside an argument get verdicts, the argument is left alone, and the attribution is checked — who said it, where, and whether the text quotes them correctly. A checker who tries to verify an opinion has misunderstood the job.
Can you write it, or do we write it ourselves?
We run the check, from the text as it stands, and the corrections belong to the author. Every verdict comes back with its source and its evidence, so the author can accept a correction, dispute it, or bring a better source the checker missed — that dispute is part of how the check works, and it is how the ledger stays honest. What the work does not do is write the claims, or improve the argument while correcting its facts.
Where is Writely available, and how do I start?
Writely is based in Nigeria and works remotely with writers, newsrooms and organisations across Nigeria and the rest of Africa. Use the Chat with us button to send the text first, then the deadline and the publication date. If the claims came from a research document or an interview, that is the most useful thing to send, because the claim register can be built from where the claims actually began.
What do I get at the end?
The claim register listing every claim extracted and decomposed, the verification report giving each one its verdict with its source and evidence, the corrected text with changes tracked to the findings, the source file with citations, access dates and archived copies of every checked link, and the re-check plan naming what decays and when to look again. The publication decision is not included — that stays with the editor.

Your next move

Send us the text, not the topic. We’ll work out what the claims can honestly stand on.

Whether you need the claims extracted and decomposed, the source hierarchy applied, three honest verdicts per claim, or the ledger that lets the text be checked again, begin a clear conversation with Writely.

Request services