How we keep spam and off-topic posts off TrustList

How we keep spam and off-topic posts off TrustList

For a long time TrustList behaved like an open noticeboard. Anyone could add a listing, write a guest post, make a public profile or leave a review, and very little stood between what they typed and what a reader saw. Some of what arrived was useful. A good deal was not: pages for vape shops and essay writers, profiles that were adverts with a name attached, guest posts that existed to carry a link back to the author's firm, and old listings whose text had been swapped for something else entirely when the business's website domain changed hands.

That has changed. On 26 September 2026 we scanned all 89,840 live listings for spam signals, had every one of the 1,719 that were flagged read in context, and took 176 spam or illicit listings and 52 promotional guest posts offline. Between 24 and 29 September we wrote down what TrustList covers and what a profile is for, held 22 advert profiles by 25 September, and took a further 13 down on 29 September. Alongside the clean-up we put the rules where they bite: at the door, where a listing or article is published, and in review afterwards.

This article is the story of that move, from a directory anyone could fill with anything to one with rules that are enforced. The wider clean-up is told in How we cleaned spam, thin pages and dead links, and the guest post rules in Our rules for guest posts and listings. Here we concentrate on spam and out-of-scope posts, and on the controls that keep them out.

What spam and off-topic posts looked like here

Two different problems were mixed together, and it helps to keep them apart.

Spam is content that is there to game a system or to sell something unrelated to the reader's purpose. Off-topic content can be perfectly honest and still not belong: a removals firm, a wedding venue or a jeweller is a real business, but TrustList compares business software, technology services and the agencies, recruiters, outsourcers and funders bought alongside them. A reader looking for a payroll tool has no use for a worktop supplier.

Here is what each looked like in our own records.

Listings whose text was not theirs

The audit on 26 September found the sharpest examples. In our full-text scan of 89,840 live listings, six reviewers read the 1,719 flagged ones in context, and sorted them like this:

• 164 listings were wholly spam, and 24 more were partly spam. The text on the page belonged to a domain that had been taken over: a recruitment agency whose description was adult listings, a marketing agency described as an escort service, slot-machine lists on venture capital firms, and a love-spell pitch with a phone number.
• 12 were illicit in themselves. Ten were essay mills and assignment writers (contract cheating is an offence in England), plus a service selling fund "recovery" and an astrologer. Nine of them had claimed their own listings.
• 432 were junk: parked domains with asking prices, "this site can't be reached" pages, placeholder text, unedited theme demos, and a privacy policy stored where the description should be.
• 1,087 were false alarms. A keyword had matched, but the listing was fine: software vendors for the gaming industry, and real firms writing in their own language. We left them alone.

Most of the spam was not the business's doing. Old listings had been copied from company websites years ago; when the domain lapsed and someone else bought it, the text on the other end changed and our copy of it went with it.

Profiles that were adverts

On 24 September the owner found a public profile that was an advert for a vape shop: sales copy for a biography, one link to a product page, and nothing else. A check of the 34 public profiles that followed found that most were businesses rather than people. Fifteen promoted businesses that are plainly out of scope: vapes, forex and prop trading, essay writing, jewellery and gift shops, a wedding venue, removals, building supplies, worktops, industrial valves and an entertainment agency. Several more used keyword-labelled links, the kind that read like a search phrase plus a city.

By 29 September the pattern had come back. Sixteen profiles were public again, and 13 of them were businesses posting themselves as people, with the company as the "name" and a slogan as the headline. Some read "Best Digital Marketing Agency In Uk"; another called itself a "#1" firm in an Indian city.

Guest posts that were link vehicles

Before 26 September there were 61 guest posts. Fifty-one of them were promotional: backlinks to the author's firm, "hire us" paragraphs, or press releases. One was spun text. We also found a piece published as editorial that was an unfinished stub, which told readers to set an access key "for a fully written article" and then listed raw scraped blurbs. That was our own failure, and we took it down with the rest.

Why spam and off-topic posts hurt readers and search

The harm to a reader is simple. A directory is useful only if what is in it is what it says it is. If one listing in a category is a parked domain and another is a taken-over site offering something a reader would not want to find, the reader cannot trust the neighbours either.

There are three further costs. Spam wastes a reader's time and can hurt them: a listing that leads to adult content, a scam or an essay mill is worse than no listing. It damages the honest businesses beside it, because every off-topic or junk entry in a ranking is a place taken from a business that earned it.

And it costs us in search. Search engines say plainly that thin, copied, automated and link-scheme content is spam, and that a site carrying it can lose visibility. Our own coverage figures showed the effect. By 21 September, 90,416 pages were indexed, against 139,511 on 25 August, and daily impressions had fallen from 22,638 on 18 August to about 4,000 to 5,000. Search engines were declining pages they judged thin.

That is why we wrote rules that can be checked.

The rules we wrote

On 24 September we found that the guidelines said nothing at all about what TrustList lists. There was no line between an agency, which is in scope, and a removals firm, which is out. So there was nothing to point to when we needed to say no. The rules below now sit in one place, on our listing guidelines and contribute pages.

Scope: what we list and what we do not

TrustList lists business software, technology services, and the agencies, recruiters, outsourcers and funders that are bought alongside them. We do not list:

• shops and retail sellers, including vapes and gifts;
• gambling, trading and loans;
• essay writing and assignment help;
• local trades and services such as removals, building supplies and venues;
• general accountancy and legal work that is not about technology;
• adult content.

There is one test for unclear cases, and three decisions in the wording that follow from it:

• Software built for any industry is in scope, but the industry business is not. A pizzeria's point-of-sale system can be listed; the pizzeria cannot.
• Accountancy, legal and certification firms are in scope only when the work is about technology: security certification readiness, technology contracts, software tax credits. General accountancy, payroll and local accountants are out.
• Where a firm does both, we list the in-scope part only.

Off-scope subject is also the first listed reason a guest post can be rejected.

Profiles: a person, not an advert

A public profile is a person, under their own name. That means:

• your own name, not a company, brand or shop name;
• a headline that is your role, not a slogan, and never "#1", "the best", "top-rated" or "leading";
• an About that is about you, not sales copy;
• links to your own site, your professional profile and your firm's listing;
• no keyword-labelled links, no links to shop or product pages, and no promotion of a business outside our scope.

If you are here for your business, the way in is to add a listing, and the free plan covers one.

Articles and guest posts

An article needs at least 2,000 words, one main heading and four or more section headings, a descriptive title of at least 20 characters, and a real author. It may not carry affiliate links or link exchanges, and claims must have a source at the point where they are made. A vendor article cannot be marked as editorial; that label is ours alone.

Titles have their own rule: "Top", "Best", "Leading", "#1" and "... in " are how our own ranking pages are named, so vendor articles may not use them.

The checks at the door

Rules that depend on someone noticing are rules that fail on a busy week. So the important ones are built to run whenever something is created or published, whichever route it arrives by. The website form, the vendor tools and any future connection all go through the same checks. Staff can bypass them for editorial work; vendors cannot.

This is what a listing or article must pass before it can go live.

For an article:

• at least 2,000 words of prose, counted without code, tags, links or web addresses that could pad the total;
• at least one main heading and four section headings;
• a title of at least 20 characters that does not read like a ranking page;
• the "editorial" label is forced back to vendor-written if a vendor tries to set it.

An article that fails stays saved as a draft, and the message tells the author its word and heading counts against the minimums.

For a company or product listing:

• a description of at least 120 characters, so a reader has something to read;
• at least one category, so the listing can be found;
• a working website address that is a proper web link rather than a placeholder;
• no more than 20 categories. A single accounting office had once carried 565 tags, mostly countries, to show up everywhere at once. Twenty is generous, since most listings carry fewer than ten.

Saving is never blocked: a half-finished listing can sit as a draft indefinitely, and the check fires only when something is about to go live. Listings already live before the checks existed are not locked out, so an owner can fix the very thing being asked for. An edit that makes a listing worse than the live version is still refused.

Review and takedown

Checks at the door catch the careless and the automated. They do not catch a listing that passes on paper and is wrong in fact, or a legitimate page that turns bad later when a domain changes hands. For that we review, and we act.

Archive, not delete

Nothing we remove is destroyed. Off-scope and spam listings are archived: they stop being shown, stop being linked and answer as not found, but they stay on record with a list of what was changed so that any of it can be undone. The same is true of advert profiles, which are hidden rather than erased. That matters because we can make mistakes, and because it lets us show what was done and when.

Reading before removing

A keyword match flagged 1,719 listings, and people read every one before a decision was made. That is why 1,087 were left alone. The 24 partly-spam listings were real businesses with injected text, so we rebuilt them instead of removing them.

The junk and partly-spam listings then went through a research queue of 456 rows, each rebuilt, retired, removed or left for a decision on the business's own current pages. On 27 September the first batch rewrote 29 and removed 23; the second rewrote 19 and removed 11.

Holds on profiles

Hiding a profile is not enough if its owner can tick "make my profile public" and put it straight back. On 25 September we added a staff hold: a hidden profile stays hidden, the owner cannot republish it, and the editor tells them why. By the end of that day 22 profiles were held, 15 hidden as adverts and 7 more borderline ones whose links were keyword-labelled sales links. On 29 September a further 13 were taken down under the same hold: 11 that advertised a business rather than introducing a person, and two that advertised businesses TrustList does not list. Three profiles that were people stayed.

We keep the take-down as a human decision. Nothing refuses a business name automatically. A person looks, decides, and records the reason.

Our red note

For claims like "#1" or "the best firm", we do not just ask people to remove them. When a profile's name, headline, About, role or company includes a ranking claim, the public profile shows our own note beside it. The note quotes the claim and says we have not verified it, that nobody is ranked on TrustList for what they say about themselves, and that our rankings come only from our own method.

The owner cannot edit or remove the note, except by removing the claim itself. Staff can add a note of their own, which the owner's editor also cannot change. We also closed a gap: the hold flags were not protected on one route, so a held user might have lifted their own hold. They are now protected.

Reviews

Seven reviews were hidden on 26 September, with ratings and trust scores recalculated afterwards. We also fixed a fault: blocked reviews had never actually been hidden, and now they no longer appear.

Stopping repeat senders

Spam is not only what gets published. Some of it arrives in the inbox first. Filing a thread as spam in our team inbox also blocks the sender in the same step, because filing alone is bookkeeping and does nothing to stop the same sender coming back next week. The block can be placed against the sender's address, or against their whole domain. Domain blocking earned its place early: two of the first eleven real threads came from one domain, and two more from another.

Two safeguards keep this fair. A domain block refuses shared mailbox providers, where one sender's spam would shut out thousands of unrelated people, and it says why. And every block can be undone with a second click.

Links that carry no reward

A lot of spam exists for one reason: a link. If a link on TrustList passed reward to whoever put it there, the incentive to spam would survive every rule we wrote. So we removed the incentive as well.

• Profile links are nofollow. A link on a person's profile tells search engines not to pass credit to it. Because of that, even a profile that slipped through gains nothing as a link vehicle, and the profile rule says so plainly.
• Sort links do not feed the crawler. On 28 September we made the sort choices on ranking pages nofollow and never prefetched. Default settings are no longer spelled out in page addresses. A ranking of 10 pages had been offering search engines some 70 web addresses that all point back to one page, and the export we were handed listed 34,786 pages marked as alternates for that reason.
• Action links stop being followed. "Write a review" and "Own this listing? Claim it" appeared on every listing, and each led to a sign-in page. The export counted 28,908 pages that redirect and 10,041 blocked from crawling. Both are now nofollow.
• Affiliate links and link exchanges are refused in articles, as above.

None of these is a penalty on an honest business. They stop links from being a reward for anything except a real recommendation.

Before and after, in numbers

Every figure below comes from our own dated records.

• Whole-catalogue spam scan, 26 September 2026: unread in bulk -> 89,840 live listings scanned, 1,719 flagged and read in context, 1,087 cleared as false alarms.
• Spam and illicit listings, 26 September 2026: live and shown -> 176 archived (164 companies and 12 products).
• Guest posts, 26 September 2026: 61 live -> 52 archived (51 promotional, one spun), plus the unfinished stub article.
• Reviews, 26 September 2026: 17 published, one a scam pitch -> 7 hidden, and blocked reviews now genuinely drop from the page.
• Junk and partly-spam listings, 27 September 2026: 456 rows in the priority queue -> every row rebuilt, retired, removed or left for a decision the same day. The first batch rewrote 29 and removed 23; the second rewrote 19 and removed 11.
• Public profiles, 24 September 2026: 34 checked, 15 promoting out-of-scope businesses -> hidden, with a rule on what a profile is.
• Profiles held, 25 September 2026: none held -> 22 held by staff, with 13 left public.
• Profiles taken down, 29 September 2026: 16 public again, 13 of them businesses posing as people -> 13 taken down, and a confirmation needed before any profile first goes public.
• Sort-link crawl waste, 28 September 2026: 34,786 alternate pages in the export -> sort choices nofollow and defaults left out of the address.

What this means for readers is plain: a listing on TrustList is now much more likely to be a real business, within the subject, with a description written for the reader, and a profile is much more likely to be a person. When a claim slips through, our own note is beside it.

For honest businesses the change is a relief more than a burden. The rules ask for a real description, a working website, a sensible number of categories and a person behind the profile. A business that meets them is no longer sharing a category page with parked domains and taken-over sites. If we take something down that should stay, we can put it back: nothing is deleted, and a reason is recorded each time. You can reach us at hello@trustlist.uk to ask for a review.

Sources

• TrustList listing guidelines
• Contribute to TrustList
• How TrustList works and why you can trust it
• Google Search spam policies
• Google AdSense programme policies
• Related on TrustList: How we cleaned spam, thin pages and dead links and Our rules for guest posts and listings

TrustList figures in this article come from our own dated records, kept as we did the work. Dates are 2026.