From a list of names to insight: how TrustList is changing what buyers get in the AI era

From a list of names to insight: how TrustList is changing what buyers get in the AI era

Until this September, TrustList was an ordinary directory. It came back after about a year offline, and around 17 September 2026 a quarter of our live listings (24,350) had fewer than 50 words of profile text. On 25 September 2026 our advertising application was refused for thin, low-value content. That was a fair verdict on a site that could tell you a company exists but little else.

Since then we have been turning a list of names into a curated catalogue. By the afternoon of 29 September 2026, 29,722 of 72,564 live company listings had been rewritten to one written standard, and about 12,000 published rankings now back our ranking pages. This article covers what that means for buyers in the AI era, and where the work stands as of 30 September 2026.

We decided to lead with our own research, to show readers articles that match what they are actually looking at, and to write about the things that have changed for buyers since AI tools arrived. We also decided what we would not do, which mattered as much as what we would. Where a figure comes from our records, we give the date it was true. Where we have no figure, we say so rather than guess.

Why a list of names stopped being enough

A list of names answers one question: who is out there? Buyers rarely stop at that question. They want to know which of those names suits a company of their size, in their country, at their budget, and whether the claims on the website can be trusted.

Two things pushed us to change. The first was our own catalogue. It is large: on 19 September 2026 it held 5,485 venture capital firms, 2,201 investors, 72 startup funding programmes, 51 investment funds and 264 staff augmentation companies, alongside many thousands of software products and service providers. A catalogue that size contains patterns that no single listing shows. The second was the state of the web. Search results and feeds are now full of text that looks like analysis and is actually a vendor's announcement rewritten in the third person. A directory that adds to that pile is not helping anyone.

So we asked what we can say that is true, useful and could only come from having gathered all of this in one place. The rest of this article is the answer so far.

We wrote about a related problem, search traffic, and what we got wrong there, in our earlier piece on our search traffic journey. We will not repeat it here.

Our own research comes first

The most important editorial decision of the past few weeks was about voice. In mid-September we looked at what we had been publishing and saw a habit we did not like. Articles opened with what a popular forum was discussing that day, quoted point and comment counts as if they proved importance, and only brought in our own data several paragraphs later, as a reply to somebody else's conversation. It made us look like commentators on other people's platforms.

The owner's instruction was direct: the site should establish its own identity through its own research. From that day the rule has been:

• Lead with what we found, counted from our own catalogue.
• Do not use point counts or comment counts as evidence. That a thread is popular is not a fact about the world.
• An outside discussion may be mentioned as context after our finding, never as the reason the article exists.
• Cite the primary source: the vendor's own page, the official filing, the programme's published terms. Never an aggregator's table.
• Say plainly what we did ourselves, such as reading price cards one at a time or reading funding programme terms off official pages.

We applied the rule retrospectively. An article published on 19 September had opened on a forum thread with a handful of points. Its substance was entirely ours, built on our board of 101 AI models and 23 price cards read one at a time. We rewrote it to open on the board and republished it the same day.

Topics are not restricted by this rule. What is out of scope is the piece that has nothing of ours in it.

Where our writing had become too narrow

Counting is a good discipline, so we counted. On 19 September 2026 the published articles on the site were spread like this:

• Legacy blog: 124
• AI news: 19
• Product news: 11
• News: 8
• Business news: 6
• Reviews: 1

Set the 124 older blog posts aside and recent editorial was essentially one beat. Everything written since the end of August drew on the same two datasets: the board of AI models and the board of new product launches. That is a narrow view of a directory with far more to say.

The fix was not to chase more news. It was to write from more of what we already own. We identified at least six sources of our own research and had been using two: the AI model board and the launch board. The others are the funder catalogue, the offshore talent rankings, the listing catalogue and its rankings, and site search behaviour.

We have adopted a simple rotation: avoid two consecutive articles from the same dataset unless events genuinely demand it. The evidence bar stays exactly where it was. Every figure traces to a source, and nothing is estimated.

Matching articles to what you are reading

The other half is showing the right article to the right reader.

Until late September, most article lists on the site simply put the newest first. Only one surface, the "Related reading" block, looked at subject matter at all, and even that sent only the first term of a listing page. A reader looking at a cybersecurity vendor could be offered this week's unrelated news because it was new.

The underlying problem was worse than the display. Of 231 live articles at the time, 183 carried no subject term and 203 carried no place. The site could not match articles to readers because it did not know what the articles were about.

On 29 September 2026 we changed both. The site now ranks articles by how many subjects and places they share with the page you are on, and only then by date. A listing page passes on all of its terms, such as its service lines, skills and countries. A city page adds its city. A category page passes its biggest subject areas, but leaves places out so that every category is not led by UK news. Archived articles no longer appear in any of these lists, which fixed a smaller bug where a rail could offer a link that led to a not-found page.

Then we tagged the backlog. A matching tool reads each untagged article's title, summary and body and links it to terms that already exist. It never creates a term or removes a link, and every run keeps a record so it can be undone. On the live site it tagged 166 articles. Afterwards, of 236 live articles, 28 had no term and 118 had no place, down from 183 and 203.

We checked the results by eye as well as by count. An article about a data breach now sits beside other breaches rather than beside whatever was published most recently.

What AI has changed for buyers and for service firms

Much of our AI coverage is deliberately practical. Buyers of software now face questions that barely existed a few years ago, and firms that sell services face pressure from the same tools.

For buyers, the new questions include:

• Which AI models are available, what do they cost per unit of work, and how are those prices quoted? Our board records models and their price cards. Reading 23 of those cards one at a time was itself a finding, because the information a buyer needs is presented in different ways from one supplier to the next.
• What does a vendor mean when it says a product is "AI-powered"? A feature and a marketing label are not the same, and we try to describe the first.
• What happens to your data? Our checklist for a data processor under GDPR runs to 17 questions, and our security review checklist to 16.

For service firms, the pressure is different: clients ask how AI changes pricing, staffing and delivery. We treat those as questions for research rather than opinion, and say when we lack the evidence.

We also use AI-assisted drafting ourselves, and you should know how. It is used to help draft listing text, under strict written rules and with automated checks on the output. It does not supply facts. A fact enters the catalogue only if it comes from a source we can name, and unsupported claims are flagged with a visible note rather than repeated. You can read the rules for contributors on our listing guidelines page, and how we think about trust on our trust page.

A fuller map of the questions buyers ask about AI features, tied to individual products, is still being built.

Funders, due diligence and the other side of the table

In September we asked ourselves a question that changed our plans: if we hold a large catalogue of software and service businesses, why do we not write for the people funding them, and for the founders looking for funding?

It turned out we already held the material. The funder catalogue is described above. Beyond the counts, it carries researched evidence for each funder: the stage it invests at, what kind of investor it says it is, sectors, offices, cheque size as the funder wrote it, founding year and an application link. Each item is kept with the sentence and page it came from. We also built 72 funding programme listings from official pages with verbatim quotes and links, covering schemes in the UK, the US, Saudi Arabia, the UAE and India, and funding guides for six countries.

There is one thing we deliberately do not have, and it shapes what we write. We do not hold round data. We do not record who raised how much in which quarter, and we do not treat a fund's size as a cheque size. So you will not find a deal round-up from us. Doing so would mean taking numbers from elsewhere and presenting them as ours.

What we can say, with counts nobody else has assembled, includes:

• how many funders state a cheque size at all, and how that compares with the stage they claim to serve;
• which stages are actually served in which cities, across 68 hub cities;
• what official programme terms actually say, in one place.

We also studied how others present this material. One widely quoted article ranking large venture firms contradicted itself: its table and its own FAQ gave different figures for several firms. We decided never to cite it, and to borrow only its good habit of showing the basis and date beside every row and saying what a ranking is not.

On 19 September 2026 we also launched a section on due diligence, aimed first at buyers checking a vendor. There are seven checklists: five for buyers (SaaS vendor, IT services provider, offshore development partner, GDPR data processor and security review) and two for founders (a pitch deck and data-room readiness). Each is printable, can be downloaded as a ticked-off list, and carries a version and a last-reviewed date.

Alongside it we built a fundraise board where founders choose what is public. Amounts, traction and identity sit behind a gate, and founder-stated metrics are shown as self-reported, apart from anything we checked.

Some honest gaps remain. Nobody has been invited to the board yet. Sector and region are free text on both sides, so "fintech" does not match "financial technology" until we back those fields with our thesaurus. The evidence that would let a checklist say what we hold about a particular vendor has not been collected yet. Publishing fundraising requests may also count as a financial promotion under UK law, and we have flagged that for legal advice rather than assume the answer.

If you run a funding programme or an investment firm, the contribute page and the page for vendors explain how to correct or claim a listing.

Offshore talent and what the catalogue can show in aggregate

The other side of the table from funders is talent. In September we built rankings of staff augmentation firms by country and by city, starting with India, Ukraine and Pakistan, with Saudi Arabia and China held back until each has enough evidenced firms to justify a page.

The facts underneath are sobering, and we published them. Ratings across the directory were close to empty: on 16 September, twelve published reviews and fourteen companies with a legacy rating. Addresses were missing for more than 99% of the firms in these countries. Country tags carried errors, with our audit finding hundreds of contradicted tags. So the rankings say what they are ordered on, and they do not invent ratings to look fuller than they are. Signals count only if a firm earns them: facts from its own site with the source kept, its answers to our ranking questions, vetting, and reviews that come through invitations.

Website research gave 770 firms an evidenced staff augmentation tag; firms whose only evidence was their own listing text were held back.

The same honesty applies to the wider catalogue. On a copy of the live catalogue taken on 19 September, of 12,116 live software products, 4,963 stated a starting price and 4,856 offered a free trial. Of 85,420 live service companies, 5,681 stated a founding year, 243 an employee count, 146 an hourly rate, 112 a minimum project size, and just five an enquiry response time.

These figures describe what our catalogue holds, not what vendors publish on their own sites, so they are a floor. But they support a plain finding: the information a buyer needs in order to compare suppliers is, for most of them, not written down anywhere we have found it.

Some aggregate work is in place: since June, rating distributions, per-dimension averages and ratings over time for each listing.

What we will not do

Being clear about refusals is part of the change, so here they are.

• We will not publish press releases as analysis. If a vendor announces something, we look for the primary source, check it, and write about it only if we have something to add. Announcement-shaped text with our name on it is exactly what we set out to avoid.
• We will not invent figures. No estimated market totals, no funding round sums, no ratings we did not gather. Blank is an acceptable value.
• We will not write from a forum. A popular discussion is at most a lead for us to check.
• We will not let sponsorship blur the ranking. Where our own listings appear, they are labelled as sponsored and they rank on what they earn.
• We will not treat AI text as evidence. Drafting help is fine. Facts need sources.

These rules cost us speed. We think that is the right trade for a site whose only product is whether you can trust it.

Before and after, in numbers

Here is where things stood before the clean-up and where they stand now, each figure with its date.

• Thin profiles: around 17 September 2026, 24,350 live listings had fewer than 50 words of profile text -> by the afternoon of 29 September 2026, 29,722 of 72,564 live company listings had been rewritten to the standard, with about 19,000 more queued to publish that day.
• Whose 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 -> on 28 September 2026, about 11,500 of 81,127 live, active listings were in our own words (14%), and by 29 September 2026 the rewrite had reached the 29,722 above.
• Spam: by 27 September 2026, 164 wholly spam and 24 partly spam listings and 52 paid guest posts had been archived, and 7 reviews blocked.
• Thin listings: 6,165 were rewritten from the business's own site, and 15,238 with nothing to source were parked on 27 September 2026 (archived, and reversible).
• Dead sites: 2,675 listings whose domain had no DNS record or was parked were taken offline by 27 September 2026, and 4,530 more were flagged so that their pages never link a dead site.
• Ranking pages: about 18,000 sat in the sitemap, many built from import tags, and 68,678 category-and-term pairs had a page although only 11,962 had five or more listings that were not thin -> on 28 September 2026 a ranking page exists only when a published ranking backs it, about 12,000 in all. Category pages without one answer 404 and nothing links to them.
• Articles and subjects: of 231 live articles, 183 carried no subject term and 203 no place -> on 29 September 2026, of 236 live articles, 28 had no subject term and 118 no place.

Every rewritten profile follows one standard: what the business offers, who it works with, and key facts. It uses only facts from the business's own site and our records, and it passes automated checks for length, copying, hype, invented numbers or names, prices and contact details.

For a reader, the change is mostly about what you no longer have to filter out. A page that exists now has a reason to exist, a profile is written by us rather than lifted from the seller's own pitch, and a ranking reflects a published ordering rather than an import tag. It does not make every listing complete, and this article says so where it matters. It does mean that when we show you something, we can name the source and the date.

Sources

Figures about TrustList in this article come from our own records, each with the date it was true: 16 September 2026 for offshore country data, 19 September 2026 for the catalogue, editorial and due-diligence counts, and 29 September 2026 for article tagging.

TrustList pages:

• Listing guidelines
• Contribute
• Trust and vetting
• For vendors
• Our search traffic journey
• Rankings

Public pages from Google that inform our editorial standards:

• AdSense programme policies
• Google spam policies
• Creating helpful, reliable, people-first content