Adverse media in due diligence

The problem

Before a deal closes, a supplier is onboarded or a distributor is appointed, somebody has to answer whether there is anything in the public record that should change the decision. That search is usually done in English, over a working day, by someone who is not a researcher, and it stops when the first two pages of results look clean. The gap is geographic and linguistic rather than one of effort: a counterparty's difficulties are reported first, and often only, in the press of the country it operates in. A Turkish construction group's payment disputes appear in Turkish trade coverage; a Vietnamese manufacturer's environmental fines appear in Vietnamese local reporting. Neither reaches English-language coverage unless the counterparty is large enough to be internationally newsworthy, and most counterparties are not.

How does a news API support adverse media checks in due diligence?

Run the check against the index rather than against a search engine, in the languages the counterparty actually operates in, over an explicit window you record. Meaning-based retrieval matters more here than almost anywhere else, because adverse coverage is written in euphemism as often as in plain terms — a company does not have its licence revoked, it 'agrees to suspend operations pending review'. A query composed in English reaches those formulations in other languages without anyone on the team composing a query in Turkish or Vietnamese. Collapse to stories so that one incident covered by thirty outlets registers as one finding rather than thirty, then read the outlet count as a rough measure of how widely the incident was reported rather than as a measure of how serious it was.

What the workflow looks like

1. Resolve the counterparty to the names it is actually reported under: registered name, trading name, the local-script rendering, and the names of the directors you are contracting with. resolve_entity returns the spellings the index holds for each; a filter is exact, and a spelling it does not hold reads as a clean record.
2. Decide the language set from where the counterparty operates, not from where you sit. This is the single decision that determines whether the check is worth running.
3. Ask related_entities which people and organizations the counterparty is most often reported alongside. An unfamiliar director or intermediary there is a question for the counterparty, since co-mention shows who appeared in the same coverage, not who is connected.
4. Run one query per name over a window long enough to cover the relationship you are contemplating, typically several years rather than several months.
5. Collapse to stories, then read the outlet spread on each finding to separate one local report from something that was widely covered.
6. Record the query, the window, the language set and the date the check ran, alongside the findings. The record of what you looked for is the part that matters six months later.
7. Route every genuine finding to a person before it reaches the decision. Nothing here should gate a contract automatically.

What will a news API not do for adverse media checks in due diligence?

This is not a screening product and must not be used as one. Sanctions, politically-exposed-person and watchlist screening are regulated activities with defined list coverage, audit requirements, match-quality standards and vendor accountability, and none of that is on offer here. What this gives you is news coverage, which is a different and much weaker instrument: it is silent on anything that was never reported, it cannot tell you whether an allegation was substantiated or subsequently withdrawn, and its entity resolution groups spellings rather than confirming identities, so a name match is a string match and the burden of confirming identity is entirely yours. If your obligation is regulatory, buy a screening vendor. If your question is commercial, this is a reasonable way to ask it.

The mistake teams make here

The mistake is treating a clean result as evidence of a clean counterparty. Absence of coverage is overwhelmingly the normal state — most companies of most sizes in most markets are never written about — so an empty result set carries almost no information, and it is psychologically very difficult not to read it as reassurance. It is worse in markets where the press is constrained, which are frequently the markets where the risk is highest: a counterparty in a country with a weak independent press will generate less adverse coverage than an identically behaved counterparty in one with a strong one. Reporting a nil return as 'no adverse media found' rather than 'nothing found, in a market where little is published' is how a check that measured press freedom ends up in a deal file as a finding about conduct.

How would you know adverse media checks in due diligence is working?

Test it against the cases you already know. Take five counterparties where something eventually went wrong and something was reported at the time, run the check as it would have run before you engaged them, and see whether it surfaces the coverage. That tells you your recall on the kind of problem you actually encounter, which is far more useful than a general claim about coverage. Then measure the other direction: count how many findings a reviewer dismissed as the wrong entity. Above roughly one in three, your name resolution is the problem rather than the retrieval, and adding query terms will not fix it. Track both numbers per market, because they diverge sharply between markets with a dense press and markets without one.

What to build in the first week

Start with the twenty counterparties you are most exposed to rather than building a pipeline for all of them. Resolve their names by hand, including local-script forms, and write the list down — name resolution is the work, and automating it before you have seen what the hard cases look like produces a system that is confidently wrong. Run the checks manually, read everything that comes back, and keep a note of every result you had to discard and why. By the end of the week you will have a realistic false-positive rate per market and a clear sense of which languages matter for your supplier base, and those two facts determine whether the automated version is worth building at all.

What are the alternatives for adverse media checks in due diligence?

Dedicated adverse-media and screening vendors maintain curated source lists, perform entity resolution against corporate registries, and stand behind a match — which is exactly what a regulated process requires and what this does not attempt. Corporate registries, court records and regulatory publications are authoritative where news is merely indicative, and for a material transaction they are the sources a lawyer will want. Commissioned investigation firms reach what is not published at all, which is often the point. Use this where the alternative is an English-language web search done in an afternoon: it is broader, cheaper, faster and reproducible, and it is not a substitute for any of the above.

The queries that implement it

• Watch what is said about a named individual — /stories in hybrid mode
• Monitor a market whose language nobody on the team reads — /stories in semantic mode
• Watch for regulatory action in a sector — /stories in hybrid mode

Each recipe makes the case for its parameters. Response fields, failure modes and scheduling cost are in the technical documentation.