What exactly does the source check do?

It runs over the article before publication and opens every source link in it. Three things come back flagged: a link that will not load, a link whose target has been replaced by a homepage, and a source old enough to be worth reading again. Each flag names the sentence it belongs to, so you fix a line instead of hunting for one. The limit matters, because your readers will find it anyway. The check tells you the link works and still points where it should. It does not tell you the source is right, and a link that resolves is not the same as a claim that holds.

Does cpywrk check whether a technical claim is correct?

No. It does not verify technical accuracy, and nothing here should be read as suggesting otherwise. The narrower thing it does test is whether the draft makes a claim the research behind it will not carry, so a sentence that went a step past its evidence is marked before you read it. Deciding whether an assertion is true stays with you, and nothing publishes until you say so.

Can we set rules about what we will never claim?

Yes, and they hold across every article instead of being retyped into every brief. No figure without a named source. No competitor named. No fear framing. Those go into the brand profile once, the brand voice check tests every draft against them, and anything you correct later joins the same list, so the standard only moves one way.

Our sources go stale faster than we republish. Does that get better?

The check looks at the article in front of it, not at your archive, so it does not repair anything you put out two years ago. What it changes is the next two years: sources are named in the text, the links are opened and tested before publication, and an ageing one is flagged against the passage it supports rather than found by a reader. If an old piece needs revisiting, bring it as a topic and it goes through the pipeline like anything else.

Your reader will follow
every link you publish

In security the audience checks things for a living. They open the source, look at the date, and notice when the page you cited no longer says what you said it said.

Written for security software companies whose readers audit claims for a living, and whose credibility is the thing being bought.

Start free trial See what changes here

One full article free. No card required, no time limit.

What changes here

Every link is checked while the article is still a draft

Your reader clicks through. Not out of suspicion about you, but because verifying a claim is what they are paid to do all day, and the habit does not switch off for a blog post. The ordinary decay of a link costs most publishers nothing. It costs you the reader. A dead citation in a security article does not read as a broken link. It reads as an answer to the question they were already asking.

So the checking happens before publication rather than after somebody writes in. While the article is still a draft, cpywrk opens every source link in it and reports what it finds: a link that will not load, a link that now delivers a homepage rather than the page you cited, and a source old enough to deserve a second look. Each one is reported against the sentence it belongs to. If nothing is wrong, the article goes out and you hear nothing about it.

Be exact about what that is, because this audience will be. The check confirms a link resolves and still points where it should. Whether the source was any good in the first place is a judgement, and it stays yours: a working link is not a true sentence. What the check takes off your desk is the failure that never needed judgement at all, the citation that stopped being true while nobody was looking at it.

A link that will not open

Caught while the article is still yours to edit, so the repair is a sentence rather than a correction notice.

A link that quietly moved

What you cited is gone and the domain now sends everybody to a homepage. In the draft the citation still looks immaculate. Same check, same flag.

A source getting old

Age is surfaced, not judged. Whether the source still holds is your call, and it is better made now than in a reply to somebody who made it for you.

One more thing runs on the draft before you see it: a test of whether it makes a claim the research behind it will not carry. That is the sentence which went a step past its evidence and got through because it sounded plausible, and it is precisely the sentence your audience is best at finding.

Where the line sits, plainly. cpywrk does not judge whether a technical assertion is true. It cannot verify technical accuracy and it is not standing in for your own read. It gives you a draft whose sources are named, whose links open, and whose claims trace back to something.

A single printed sheet held up to a strong backlight so the text blurs, one line on it circled in violet, a hand holding the edge of the page.

The work itself

Fear ages badly. An explanation does not.

The instinct in this market is alarm, and it works for about a week. Then the incident is old news, the urgency has expired, and the piece sits on your site with a date on it, ageing in public in front of the one audience that reads dates. Nobody links to a warning. They link to the thing that explained it.

The articles still earning links in two years explain a mechanism. How the thing actually works, in order, including the step everyone skips. What a control does and, more usefully, what it does not stop. Why the obvious fix holds in a test and falls over in a network somebody has to keep running. They are also what an assistant reaches for when somebody asks it how to handle the problem, because an explanation survives being lifted out of its context and a warning does not.

Take the mechanism apart

Start to finish, in order, with nothing waved past. It is the hardest thing to write and the only thing still being read in two years.

Name what it does not stop

A control described with no limits reads as marketing, and this reader stops there. State the boundary and the rest of the page gets believed.

Show where the number came from

Nothing invents a figure. A number traces either to the brief you approved or to a named source sitting beside it in the text.

Leave the alarm out

Explaining something properly removes the need to frighten anybody about it. Write that into the brand profile once and every draft is held to it.

Corrections behave the same way. Strike a phrase once and it stays struck, so the standard climbs rather than resetting with every article. None of this wins a quarter. It settles the only question your reader is really asking, which is whether the things you publish hold up when somebody goes and checks them.

Where it ends up

A finished article still has to get onto the site

Credibility in this market sits on your own domain. That is where somebody goes to work out whether you know what you are talking about, and an article that never reached the site proves nothing to anyone. Unpublished work is not cautious. It is invisible.

Webhook, into your own build

The finished article, its metadata and its images are handed to whatever builds your site, so publishing stops waiting for an engineer to have a free afternoon.

WordPress, direct

If the marketing site runs on WordPress, connect it once. After that it is one button: press Publish and the article is live, formatted.

Publishing can also run on a schedule you set: a date and time per article in your own time zone, or a batch at one or two a day. Any scheduled article can be rescheduled or cancelled.

The LinkedIn, X and Instagram versions come out of the same brief, written for how each one reads. Those are yours to post when you choose. We write them; we do not post them for you.

Before it reaches you

Six checks, every article, before anyone reads it

In a market that sells trust, being caught approximating costs more than a quiet quarter. So the grading below happens first, and what lands in front of you has already survived it.

Search

It targets a question people actually type, and it is built to rank for it.

AI answers

Buyers now ask an assistant to draw up the shortlist. The article is structured so those systems can quote it.

Brand voice

It sounds like your company, checked against rules you set once rather than re-explained every time.

Machine tells

Hedging, filler and the giveaway phrasing are removed before the draft reaches you.

Brief compliance

It covers what the brief said it would, rather than drifting somewhere easier to write.

Learnings

Corrections you made last month are applied this month, so you stop repeating yourself.

Before a draft reaches you it has also been checked for claims the research does not support, and every source link has been checked live. See every step up close.

You approve the outline before writing starts and read the draft before it publishes. Nothing goes onto your site that a person has not said yes to.

Proof

Read something it made and judge for yourself

We are not going to show you another vendor's traffic chart. What we can show you is our own work. Every article on this blog came out of the same system this page is describing: researched, written, checked, and published, then read and approved by a person before it went out.

The subject is content and search rather than your category, because that is what we sell. Read one anyway and answer the only question a sample can settle: is this the standard you would put your company's name on?

Read the blog

Questions worth asking first

What exactly does the source check do?: It runs over the article before publication and opens every source link in it. Three things come back flagged: a link that will not load, a link whose target has been replaced by a homepage, and a source old enough to be worth reading again. Each flag names the sentence it belongs to, so you fix a line instead of hunting for one. The limit matters, because your readers will find it anyway. The check tells you the link works and still points where it should. It does not tell you the source is right, and a link that resolves is not the same as a claim that holds.
Does cpywrk check whether a technical claim is correct?: No. It does not verify technical accuracy, and nothing here should be read as suggesting otherwise. The narrower thing it does test is whether the draft makes a claim the research behind it will not carry, so a sentence that went a step past its evidence is marked before you read it. Deciding whether an assertion is true stays with you, and nothing publishes until you say so.
Can we set rules about what we will never claim?: Yes, and they hold across every article instead of being retyped into every brief. No figure without a named source. No competitor named. No fear framing. Those go into the brand profile once, the brand voice check tests every draft against them, and anything you correct later joins the same list, so the standard only moves one way.
Our sources go stale faster than we republish. Does that get better?: The check looks at the article in front of it, not at your archive, so it does not repair anything you put out two years ago. What it changes is the next two years: sources are named in the text, the links are opened and tested before publication, and an ageing one is flagged against the passage it supports rather than found by a reader. If an old piece needs revisiting, bring it as a topic and it goes through the pipeline like anything else.

A different kind of software

Or start from what content has to do for any software company.

Run one article on your own site and see

You bring the topic, approve one outline and read one draft. The research, the writing, the checks and the publishing happen without you, and if the draft misses, you send it back: revisions are unlimited and cost nothing from your plan.

Start free trial