How we write a TrustList listing profile now, and what we check before it goes live

How we write a TrustList listing profile now, and what we check before it goes live

A TrustList listing used to be whatever the business, or an old data import, had put into the box. Around 17 September 2026, 24,350 live listings, about a quarter of the catalogue, carried fewer than 50 words of profile text. Of the older company profiles with 50 words or more, 82% (54,541 of 66,473) were written in the business's own voice, with "we" and "our". We were an ordinary directory: we repeated what each business said about itself.

Now every rewritten profile is written by us, in the third person, to one standard, and checked before it goes live. On 28 September 2026 about 11,500 of 81,127 live, active listings were in our own words, 14%. By the afternoon of 29 September, 29,722 of 72,564 live company listings had been rewritten. This article explains the standard in plain terms: what a profile contains, where the facts come from, what we refuse to write, the checks a draft must pass, and what happens when a business's website tells us too little. It also says, without softening it, that the drafting is AI-assisted, and describes the human rules and automated checks around that.

We are also telling you how much is done and how much is not. Figures about TrustList in this piece come from our own records, and each carries the date it was true.

Why we changed how listings are written

In late September 2026 an advertising review told us our site was not ready to show ads, citing low-value content. We do not think that was unfair. When we counted, we found that about a quarter of live listings carried fewer than fifty words of profile text. Of the rest, much was the firm's own website copy, copied long ago.

A directory that repeats what every business says about itself adds nothing a reader cannot get from the business's own page. So the question we set ourselves was narrower and more useful than "how do we pass a review": what would a reader actually want from a profile, and can we say it in our own words, truthfully, for tens of thousands of businesses?

The owner's instruction, in his words from 28 September, was that historical and recently added listings should be brought up to our standard, and that thin pages Google lands on should either be improved or removed. Contact details such as email, phone and location were set aside as a separate, later round. Archived listings were left alone; the live and active ones came first.

We have written separately about how search traffic has behaved for us, in our search traffic journey, and we will not repeat it here. This piece is about the profile itself.

The standard profile

Every rewritten profile has the same shape, so a reader who has seen one knows where to look on the next.

• A plain tagline. Between 8 and 20 words. It says what the business is, not how wonderful it is.
• An opening paragraph that names the business. The first sentence says who they are and what they do, so the page makes sense to someone who lands on it from a search.
• A section on what the business offers. Three to six bullets, each starting with a bold lead word or phrase, so it can be scanned.
• An optional section on who it works with. Included only when the business's own material says so.
• Key facts. Only facts the business or our records actually state, such as a year founded, a place or a named ownership.

Length is deliberate. A normal profile runs from 150 to 300 words. When the sources are thin, the brief allows 110 to 150 words, on the principle that a short, true profile is better than a long, padded one.

The style rules are as important as the shape:

• Third person throughout. The profile reads as an independent writer describing the business, not as the business talking about itself.
• British English.
• No promotional language. No "trusted partner", no "solutions", no claims of being the biggest or the most experienced.
• No prices or money amounts. Prices change, are often unstated, and were among the least reliable things in the old catalogue.
• No contact details in the text. Those live in structured fields, and their collection is a separate round.
• Nothing copied. We do not lift sentences from the business's site.

The public page shows the section headings and bullets in a consistent layout. Rebuilt listings also carry a line saying the profile was written by TrustList from the business's own published information, with the month.

Where the facts come from

There are two sources, and only two: what we already held about the business, and the business's own website, read on the day the profile is written.

For each site, a reader program fetches the home page once. If that is short, it also fetches the "about" page. If we hold no email address or phone number for the business, it looks at the contact page as well. It keeps a compact digest of what it found, and a verdict on the site: readable, dead, empty, moved or parked.

Some sites draw their content with scripts, so a plain fetch comes back nearly empty. For those we render the page in a headless browser before deciding it has nothing to say. Some sites sit behind bot walls. We do not bypass those. A business that only ever shows a wall to automated readers is treated as unreadable, and we say so rather than guessing.

Two consequences follow from limiting ourselves to those sources:

• If a fact is not in the digest or in what we held, it does not go in. The writer is not allowed to fill a gap with something that "probably" holds. A profile that says less is fine.
• We do not use third-party gossip as a fact about a business. If a reseller's page says a product has ended but the business and its owner say nothing, we do not retire the listing on that basis. In our daily thin-listing batches we have repeatedly left such cases for the owner to judge instead. Retirement needs the business's own statement, or an official record we can name.

The same discipline applies to changes of ownership and names. In the batches we ran by hand, a renamed business was renamed on our side only where the business itself said so, and an acquisition was described in the words of the parties. The old name stays in the text so people searching for it still arrive somewhere sensible.

AI-assisted drafting under rules, not on trust

We will say this plainly: the profiles are drafted with AI writing tools. A person does not type each one by hand, and we are not going to imply otherwise.

What matters is what surrounds the drafting. A writer, working on a batch of about 25 listings at a time, is given the standard, the stored text and the site digest for each business. It is told to write to the shape above and to skip a listing with a reason code when it should not be written at all. Then a checking program, which is ordinary code and not an AI, examines every draft. The writer must run that check, fix what it rejects, and may try at most twice. A draft that still fails is not published; it goes onto a list of failures for a person to deal with.

So the split of responsibilities is this. AI tools produce candidate wording. Fixed rules decide what a candidate may say. Automated checks enforce the rules on every single draft, at the volume we work at, where a person reading each one would miss things or take a year. And people set the rules, read samples, handle the failure list and answer for the result.

Nothing is published that fails the checks.

The checks every profile passes

The exact checker has changed a little between our earlier fast pass and the wider rewrite, but its core has stayed the same. A draft is refused if any of these hold:

• Wrong length. A body outside the allowed word range, whether padded or too bare to be useful.
• First person. Any "we" or "our" that would make the text sound like the business rather than an independent description.
• Promotional words. Hype vocabulary from a list we maintain.
• Prices. Currency symbols, and amounts spelled out in words.
• Numbers or proper names that are not in the source. This is the check against invention. If the draft mentions a figure, a founder or a client that the digest does not contain, it fails.
• Copied runs. A window of twelve words in a row that matches the source text is treated as copying and rejected.
• The business's name missing from the opening. The first paragraph has to say who is being described.
• Contact details in the body. Email addresses and phone numbers belong in their own fields.

Applying a batch is a second gate. Before anything touches the live site, we rehearse the whole batch on a local copy of the live database. That rehearsal catches problems the checker cannot, such as a value too long for its field, which the strict local database refuses but the live one would silently cut short. Only then does the batch go to production, and every change is recorded in a manifest so it can be rolled back. Every rewrite also records when it was made, which is how we tell the rebuilt part of the catalogue from the untouched part.

Afterwards we spot-check the live pages. None of this makes a profile perfect; it makes one that follows the rules, and a failure list that shows where the rules were not met.

When a site says too little, or belongs to someone else

A writer is allowed to decline. This is one of the most important parts of the standard, because a directory that must write something about everyone ends up inventing things. The brief gives four skip reasons:

• Not enough. The site, and what we held, do not support even a short honest profile.
• Unrelated. The website is not this business, or is about something else entirely.
• Off scope. The business is outside what TrustList covers.
• Closed. The business's site says it has closed.

In the wider rewrite, as at 29 September 2026, writers skipped 2,325 listings under those codes: 1,700 for not enough to go on, 319 unrelated, 183 closed and 123 outside our scope. Those listings keep their old copy for now. Our plan is to decide, code by code, what happens next: closed businesses become retired listings with a note explaining it, and the unsupported or unrelated ones are taken offline in a reversible way. The rule for those is the owner's call and had not been settled when this was written.

The earlier fast pass on thin listings shows what this looks like at scale. It started from 25,018 live thin listings, and by 27 September 2026, 6,165 had a new profile written from the business's own site. For the rest, we classified why we could not write one:

• No website on the listing at all, the largest group.
• A website that does not answer.
• A site with no readable content even when rendered.
• A website that is not actually this business.
• A domain that now belongs to someone else.
• A parked or for-sale domain.
• A site with too little on it to describe the business.

Those were taken offline, 15,238 in all, in a way that can be reversed. They now return a "not found" page rather than an empty or misleading profile. We deliberately did not park IT service providers during the current experiment, claimed listings, or any listing with reviews.

We also learned from mistakes. One rule wrongly parked 104 businesses whose websites had moved to a new domain after a rebrand, because their first draft had failed the checker and the rule tested the wrong thing first. We restored them.

A related check covers the rest of the catalogue: does the listed website still work? For those where the domain does not exist or is parked, we took 2,675 listings offline. For 4,530 others, we kept the listing but flagged the website, so the page no longer links to a dead or parked address and shows it struck through with the date checked. We left alone domains served through a large content-delivery network, because a bot wall there looks like a failure and is not one.

Where this stands today

As at the afternoon of 29 September 2026, 29,722 of 72,564 live, active company listings had been rewritten in our own words, and 238 of 8,561 product listings. About 19,000 more company listings were still queued to be published when we wrote that record.

That means a large share of the catalogue is still in its old form. We would rather say so than let a reader assume every page they see follows the standard. The line on a rebuilt page saying it was written by TrustList from the business's published information is how you can tell.

What is still to do:

• Finish the remaining companies, roughly 13,000 listings.
• Rewrite products, since the first pool was mostly companies.
• Bring in the IT service providers, about 14,000, which we are holding until 13 October 2026 so an experiment on rankings is not spoiled.
• Decide what happens to the 2,325 skipped listings.
• Rewrite by hand the 89 that still failed the checker after two fixes or were refused when applied, and build a draft-aware path for the 472 that have an unpublished draft row.
• Add a second layer to the fast-passed profiles, with what only we have, such as ranking placement and alternatives.

Contact details, the moved-business note showing a former name, and a fresh advertising application all come after that.

How a business can correct or claim its listing

If a profile is wrong, tell us. We would rather hear it than leave an error standing. The route depends on what you want.

• To correct a fact, use the correction route on the contribute page. Say what is wrong and point us to where your own site or an official record says otherwise. We use the same evidence standard in reverse: a source we can read.
• To claim your listing, see our page for vendors. A claimed listing becomes the business's own space. Claimed listings are not part of our batch rewrites, and we do not take them offline; we ask the business to complete the profile instead.
• If your site is mistaken for someone else's, or a domain has changed hands, tell us. That is one of the failure modes we most want to find, because it means a profile is describing the wrong company.
• If you have closed, say so. A retired listing keeps a short notice and leaves every ranking, which is what our listing guidelines promise.

Two things we will not do. We will not let a business edit the red notes we attach when a claim such as "number one" has not been independently supported. And we will not publish a profile that fails the checks because a business would like something stronger said about it.

What we learned, and what we will keep doing

Five lessons from writing tens of thousands of profiles this way.

1. Skipping is a feature. The listings the writers refused are the ones where an honest profile was not available.
2. Rules beat reviews at volume. Automated checks on every draft caught what sampling would have missed.
3. Rehearse before you publish. The local copy found problems the live system would have hidden.
4. Say what is unfinished. Most of the catalogue is not yet rewritten, and the page tells you which is which.
5. Facts come from the business or a record, not from hearsay. That costs us some completeness and buys us trust.

Our wider approach to evidence is on the trust page, and you can browse the finished work in our rankings. If you find a profile that breaks a rule described here, please tell us. That is how the checker gets better.

Before and after, in numbers

The site came back after about a year offline, and on 25 September 2026 our AdSense application was refused for thin, low-value content. This is where profiles stood before that refusal, and where they stand after the work described above.

• Thin profiles: around 17 September 2026, 24,350 live listings had fewer than 50 words of profile text -> by 27 September 2026, 6,165 of the 25,018 live thin listings had a new profile written from the business's own site, and 15,238 with nothing to source were parked, which can be reversed.
• Voice: around 17 September 2026, 82% of older company profiles with 50 or more words (54,541 of 66,473) were written in the business's own voice -> by 29 September 2026, 29,722 of 72,564 live company listings had been rewritten in the third person to one standard.
• Share in our own words: on 28 September 2026, about 11,500 of 81,127 live, active listings, or 14% -> by the afternoon of 29 September 2026, 29,722 company listings and 238 of 8,561 product listings, with about 19,000 more company listings queued to publish that day.
• Dead websites: before 26 September 2026, pages could link to a domain with no DNS record or a parked one -> by 27 September 2026, 2,675 such listings were taken offline and 4,530 were flagged so their pages never link a dead site.
• Spam: before 26 September 2026, spam sat among the listings -> by 27 September 2026, 164 wholly spam and 24 partly spam listings and 52 paid guest posts were archived, and 7 reviews were blocked.
• What a draft must pass: profiles were mostly whatever had been imported -> by 29 September 2026, every rewritten profile is checked for length, copying, hype, invented numbers or names, prices and contact details, and a draft that fails is not published.

Some of these pairs describe the same work from different angles, and the totals are measured on different days against different pools of listings, so they should not be added together.

For a reader, the change is simple to state. A profile that follows the standard tells you what the business offers, who it works with and a few key facts, and it uses only facts from the business's own site and our records. It is not the business's own sales copy, and it does not link to a website that no longer exists. A short profile now means the sources were short, not that we stopped looking. A page that we could not support honestly is no longer served at all. We have not finished, and the line on a rebuilt page saying it was written by TrustList is how you can tell which pages have been through the process.

Sources

• TrustList: Listing guidelines
• TrustList: Contribute
• TrustList: How we build trust
• TrustList: For vendors
• TrustList: Our search traffic journey
• TrustList: Rankings
• Google: AdSense programme policies
• Google: Spam policies for Google web search
• Google: Creating helpful, reliable, people-first content

Figures about TrustList come from our own records, dated in the text: counts of rewritten, skipped and parked listings are as at 26 to 29 September 2026.