Your Coverage Archive Just Got a Price: What the Ross Ruling Changes for Philippine Agencies

A US appeals court has refused, for the first time, to accept fair use as a defence for training a commercial AI product on another company's published archive. For Philippine agencies, that turns a coverage archive into an asset with exposure in both directions, and makes media content licensing Philippines, starting with a monitoring vendor's data provenance, a procurement question rather than an IT one. The ruling is narrow, still being appealed, and it does not bind Philippine courts.

Ask an agency MD what the firm owns and you will hear about people, relationships, perhaps a client list. The coverage archive almost never makes that list. It sits in a shared drive and six years of monthly reports, treated as the exhaust left behind after the real work of winning the placement.

That assumption got more expensive on 29 September 2026. A US federal appeals court upheld Thomson Reuters' AI copyright win against the now-defunct legal AI startup Ross Intelligence, in the first federal appeals case to reject a fair-use argument for using publisher content, without permission or compensation, to train a commercial AI product. What mattered was whether the finished product competed with the material it learned from. Two consequences then land on the same desk in Manila, and both are about media content licensing Philippines rather than a copyright-lawsuit scare story. Let's take the ruling first, because the detail is where the leverage sits.

What the court decided, and what it pointedly did not

Judge Tamika Montgomery-Reeves of the 3rd U.S. Circuit Court of Appeals in Philadelphia wrote that the "undisputed evidence reflects that Thomson Reuters's materials possess a creative spark, and ROSS aspired to be a direct competitor by using them for a highly similar purpose." Thomson Reuters had argued that Ross copied thousands of Westlaw headnotes, the short summaries of the legal issues in a case, to train a rival AI-powered legal search engine.

She then fenced off the question most people actually want answered. A law-firm reading of the unsealed opinion quotes the court's own line, "Unlike the AI models in Bartz and In re: OpenAI, ROSS's AI platform cannot generate original expression," and notes that the opinion "does not answer the larger questions surrounding generative AI because the technology before the court did not generate new expression." Ross's engine retrieved existing passages from judicial opinions. It wrote nothing.

Two details matter before anyone quotes this at a client. Thomson Reuters filed in 2020, long before ChatGPT, so the precedent predates the generative wave rather than reacting to it. And the holding is not settled law: the Third Circuit affirmed partial summary judgment, and on 2 October ROSS said it would seek Supreme Court review, with Yar Chaikovsky, partner at White & Case, saying the company intends "to seek review by the Supreme Court to obtain much-needed clarity on these issues." Authors Alliance, which argues the court got fair use wrong, rates the odds of rehearing en banc or certiorari as low. Stands, and is being appealed. That is enough to act on.

"Market substitute" is the phrase that matters if you hold an archive

The sentence worth copying into your own notes came from the original district ruling, which found that "the effect on a potential market for AI training data is enough." The market for an archive does not have to exist yet for its value to count against whoever helped themselves to it.

That market does exist now, and it has reported prices. A published archive behaves less like a filing cabinet and more like a back catalogue in music: it earns when somebody else wants to build on it. Press Gazette's tracker of publisher AI deals lists News Corp and Meta at up to $50 million per year for at least three years, the New York Times and Amazon at $20 million to $25 million per year (per The Wall Street Journal, itself a News Corp title reporting on its own parent), and News Corp with OpenAI reportedly worth more than $250 million over five years. Read those as reported ranges rather than disclosed contract values, which is how Press Gazette frames them too.

The Asian comparators are the ones a Philippine reader should keep. OpenAI's arrangement with Bennett, Coleman and Co, publisher of the Times of India, could be worth around $5 million per year, and one with the Indian Express Group around $3 million per year. An order of magnitude below the News Corp figure, and still a price where there used to be none.

No figure for media content licensing in the Philippines exists to quote. What the Philippines has instead is scale: a creative economy that has reached P2.12 trillion, or 7.6% of GDP, with P427 billion in creative-services exports. That is a sector-size figure and not a licensing price. It describes the body of material whose rights status nobody in a Philippine monitoring procurement has yet been asked to check.

The question nobody has put to your agency yet

Follow the logic to its uncomfortable end. If training on someone else's archive to build a substitute is not fair use, the archive is an asset, and that cuts both ways. The coverage library your firm has accumulated for clients carries value and exposure. So does every report you resell, which stands on sources you either are or are not entitled to use.

A monitoring corpus is a supply chain, and most agencies have never audited theirs past the login screen, which is what media content licensing Philippines actually means for a firm that resells coverage. The distinction between monitoring, intelligence and analytics is now partly a question about inputs: the same dashboard looks identical whether the Philippine text behind it was licensed, crawled, or quietly inherited from an aggregator. A vendor that cannot say where its Philippine corpus comes from is selling an unpriced liability alongside the interface.

The mechanism by which free content stops being free is already being sold into this market. Cloudflare pitched its crawler controls, including blocking, allow-lists and a pay-per-crawl model in testing, directly to Filipino publishers and creators in September 2025. Its own platform telemetry from Cloudflare Radar, covering January to August 2025 and not specific to the Philippines, put news and publications as the most heavily AI-crawled of 14 tracked industries, with 65% of detected AI crawlers collecting for training and 21% for search. The referral ratios in that vendor data explain publisher temper: roughly one visit back to the source per 200 crawls for OpenAI's ChatGPT, and one per 5,400 for Anthropic's Claude. Kenneth Lai, Cloudflare's ASEAN vice president, put the pitch plainly: "We are building capability to help companies to block these attempts so that we return the control back to the content creators." No named Philippine title has said publicly whether it switched them on.

Practitioners increasingly describe discoverability as presence across AI answers, social search and user-generated content rather than a blue-link ranking. Jacqueline Loch, general manager at BetaKit, argues the sweet spot is content optimised for SEO and cited in agentic search. Traci Ruether, content marketing director at Tinuiti, goes further: search, AI and social have converged, AI relies on user-generated content as training data, and treating any of them as click-through distribution channels misreads what they now are. The same compilation warns that focusing only on visibility becomes its own problem. So agencies are being asked to report on visibility inside AI answers using corpora whose rights status nobody has examined, at the moment the underlying content acquires a price. The input question belongs in the same conversation as source coverage across Philippine media.

What media content licensing looks like in Philippine practice [PH]

Nothing in the Ross decision binds a Philippine court, and nobody should let it be quoted as though it did. The Philippine test does look structurally familiar, which is what makes the market-substitution logic at least arguable here. Section 185.1 of the Intellectual Property Code (Republic Act No. 8293) provides that "the fair use of a copyrighted work for criticism, comment, news reporting, teaching including multiple copies for classroom use, scholarship, research, and similar purposes is not an infringement of copyright," assessed on four statutory factors: the purpose and character of the use, including whether it is commercial or nonprofit educational; the nature of the copyrighted work; the amount and substantiality of the portion used in relation to the work as a whole; and the effect of the use upon the potential market for or value of the work.

Read the fourth factor next to the US reasoning and the parallel is obvious. Read the rest of the provision and the gap is just as obvious. Section 185 expressly contemplates decompilation for interoperability and says nothing about text and data mining, machine learning, or AI training, because the consolidated Code dates from 1997 and has been amended only up to Republic Act No. 10372, in force 22 March 2013. [PH]

The one live Philippine instrument declines the question outright. IPOPHL's Bureau of Copyright and Related Rights put draft guidelines on AI-related works out for initial public consultation in September 2026, sorting works into four categories (AI-assisted, AI-enhanced, hybrid, and purely AI-generated, the last being ineligible) and protecting only expression attributable to human creative contribution. Registration, the draft says, covers only the copyrightable computer program and not "its training data, materials used as training data, AI outputs or other material outside the copyrightable program," and "would also not determine whether the training data or practices used to develop an AI system constitute fair use or are otherwise lawful." Director General Teodoro C. Pascua described the guidelines as clarity for registration rather than an attempt to regulate AI generally. Consultation remains open.

So the legal position on media content licensing in the Philippines is an absence, deliberately marked. What follows is analysis rather than reported fact. No Philippine publisher or media group has gone public with an AI licensing deal, a crawler block, or a formal objection, which reads as an absence of public evidence rather than proof of inaction. That silence is an agency's near-term leverage and its medium-term risk, because archives get priced eventually and contracts signed today will outlive it. The thinnest position of all probably belongs to regional and vernacular outlets, whose archives are least likely to have a counterparty negotiating for them. Copyright is also only half the exposure: the Data Privacy Act (Republic Act No. 10173) governs the personal data inside a social-listening corpus regardless of who owns the copyright in the posts.

The provenance conversation to have this quarter

Getting ahead of media content licensing in the Philippines does not need a legal budget. It needs four questions and a contract clause, which is a better use of a quarter than waiting for a Philippine test case. Ask a prospective or incumbent vendor:

• Where does the Philippine text in the corpus come from, by category: licensed directly, supplied by an aggregator, or crawled?
• What does the licence let you do? Redistributing coverage inside a client report is a different permission from internal reading.
• Are the product's AI-generated summaries derived from licensed full text or from scraped text?
• What happens to our historical archive if a publisher withdraws? A corpus that silently loses five years of a broadsheet will quietly break your own trend lines.

Then look at your own paper. Client contracts should say who owns the coverage archive and report corpus the engagement produces, and whether the client may reuse either afterwards. If you white-label a vendor's data under your brand, you are the one making the representation to the client, so the warranty needs to reach you first. And retire the oldest assumption in the business: that a clipping is a free-floating artefact rather than somebody's copyrighted work in a new wrapper. How monitoring is changing in the AI era is partly a story about citations replacing clippings, and citations travel with provenance attached.

A firm that can tell a client exactly where its monitoring data comes from is selling diligence as well as dashboards, which is a sharper pitch at the moment when every other PR agency in the market is promising AI visibility reporting. Ask your vendor those four questions. Ask us the same ones.

Key takeaways

• Cite the Ross holding as the first appellate refusal of fair use for AI training on publisher content, never as settled law, and never as covering generative AI, which the court explicitly set aside.
• Published archives now carry reported prices, from up to $50 million a year for News Corp and Meta down to roughly $3 million to $5 million a year for the Indian deals. Use the India range as the regional analogue; no Philippine figure has been reported.
• Philippine law has not answered the training-data question. Section 185 of RA 8293 offers a familiar four-factor test, unamended since 2013, and IPOPHL's September 2026 draft says registration settles nothing about whether training-data use is lawful.
• Treat media content licensing in the Philippines as a procurement question, not an IT one. The four questions above belong in your next vendor review and in your client contracts, alongside the Data Privacy Act exposure in any social corpus.

About Media Meter

Media Meter powers monitoring and measurement for Philippine agencies, white-labelled under your brand, so your team spends mornings on client work rather than clippings. Ask us the four provenance questions above, and we will answer them in a contract rather than a sales deck.