Reach plc and IBM have jointly launched a brand-safety artificial-intelligence product designed to reduce the number of news pages wrongly excluded from advertising campaigns by automated keyword blocklists, according to Campaign UK. The product enters a market where keyword lists remain the default control on most brand-safety setups despite well-documented false-positive rates.

What does the AI do?

Keyword blocklists work by string match. An advertiser uploads a list of words; any page containing one of those words is automatically removed from the ad buy. The technique scales because it requires no semantic analysis and no integration — just a list and a server-side filter. The downside is precision. A list containing "gun" will also strip ads from articles about gun-control legislation, hunting licences and Olympic shooting events. Publishers have argued for years that the cost falls on them, not on the advertiser.

The Reach-IBM system applies machine-learning classification to a flagged page before the blocklist decision executes. The model reads the surrounding sentence and paragraph context, not just the keyword token, and returns a pass-or-block verdict. A page about suicide-prevention helplines, for example, would clear the filter even if "suicide" appears on the advertiser's blocklist.

Who runs it first?

Reach is the launch publisher. The company operates a UK news portfolio spanning national tabloids, mid-market titles and regional papers, including the Daily Mirror, Daily Express, Daily Star, Manchester Evening News and Liverpool Echo. Reach's commercial team will route flagged pages on its own domains through the new classifier. IBM will take the product to other publishers and to media agencies. Neither company disclosed pricing, deployment dates or technical specifications in the announcement reported by Campaign UK.

Why now?

Brand-safety automation took hold in the mid-2010s after advertisers pulled spend from major news properties following high-profile adjacency incidents. Verification vendors — DoubleVerify, Integral Ad Science, Oracle's Moat — built page-level classification products that promised better precision than raw keyword lists. Keyword blocklists survived because they are free and require no platform integration. Reach and IBM are selling into a category where the incumbent solution costs the buyer nothing, even when it underperforms.

What we still don't know

The Campaign UK report leaves several operator questions open:

• Whether advertisers must opt in to the new classifier, or whether publishers can route flagged pages through it unilaterally
• Training data, audit trail and explainability documentation
• Latency impact on ad-server response time
• Pricing tiers for non-Reach publishers

Neither company had published technical specs at the time of Campaign UK's report.

Bottom line

Reach brings inventory and a publisher-side credibility problem to the table. IBM brings enterprise sales channels and a relationship with brand-safety platform vendors that keyword-list buyers already use. Whether the tool displaces keyword lists at scale will depend on agency holding companies and verification vendors accepting a substitute for the cheapest control they currently run.