What is scored
Earned media only: third‑party news via GDELT, queried through Google BigQuery, over a trailing 90‑day window. Owned channels and syndicated press releases are excluded. Coverage is restricted to a validated allowlist of credible outlets: national and international mastheads, US regional dailies and broadcast, and sector trade press. Stock‑tip sites are not on the allowlist, though their articles syndicated through allowlisted portals such as Yahoo Finance are scored where they name the company. Market coverage that runs in an allowlisted outlet is scored: 4% of the rows are market and ticker pages, and 89% of them carry the Performance driver, which is where a company’s financial story belongs. They are scored like any other article and never shown as evidence for a finding, because a screener page is not a story.
One story, counted once
A press release or a wire story can run in forty outlets. Counting each placement as its own article measures a company’s distribution, not its reputation, and it is uneven: on this board 19% of placements are duplicates, but the figure runs from 0% for one company to 68% for another. Syndicated copies are therefore folded into a single scored story, the highest‑credibility copy kept, and the pickups’ reach added to that story’s. A story carried by forty outlets counts once and is worth what it reached. Every card shows the placement count behind the stories, and a collapsed story carries a ×n badge.
The floor still counts placements. A brand needs 100 placements to be published, and that test asks whether enough coverage was caught to say anything at all, which folding does not change. The finer guard is the driver floor: no driver is published on fewer than 12 distinct stories.
What counts as a company’s coverage
Some companies are matched through a chain or a product rather than the corporate name: Chevrolet for General Motors, Jeep for Stellantis, Taco Bell for Yum Brands, Facebook for Meta. Those aliases reach real coverage, and they also reach a great deal that is not coverage of the company at all. An article reaches a parent through a sub‑brand only when the parent or that sub‑brand is what the article is about, named in the headline or the slug; a mention of the corporate name anywhere in the lead counts as it always has. Applying that removes 57,743 articles that would otherwise be scored against six companies. Among the kinds of story it catches are a restaurant‑inspection round‑up filed as Yum Brands and a road‑accident report filed as General Motors. The sub‑brand list is published: General Motors (Chevrolet, Cadillac), Stellantis (Jeep, Chrysler, Dodge Ram), Volkswagen (Audi), Meta (Facebook, Instagram), Yum Brands (Taco Bell), Gap (Old Navy, Banana Republic).
Page furniture is not coverage. Some outlets print a company’s name on every page they publish: a “follow us on Google News” line, a box of trending stock tickers. GDELT reads the whole page, so every article from that outlet arrives filed under the company, whatever it is about. Where a company is attached to an implausible share of everything one outlet published (at least a quarter of it), and most of those articles name it no more often than the page element itself does, with a headline about something else, then from that outlet only the articles whose headline names the company are scored for it. The test runs on each outlet and company pair separately, so a specialist outlet that genuinely covers one company is untouched. This edition it removed 11,100 articles across 21 outlet and company pairs, touching four companies. The rule is new this edition.
One passing mention is not coverage. The same furniture also turns up spread thinly across many outlets: a related‑stories link, a shopping widget, a sign‑off. An article that names a company exactly once, after the opening thousand characters, under a headline about something else, is almost never about that company, so it is not scored for it. An article that names the company in its headline, in its opening, or more than once is kept, so a company is still scored on stories it is part of without being their subject. This edition that removed 50,579 articles, 15% of everything matched to a published company. The rule is new this edition.
No topic filter. Earlier editions kept, in four sectors, only the articles GDELT had filed under that sector’s topics. That dropped coverage unevenly: a sportswear company’s coverage is filed under sport, a restaurant chain’s under food, a carmaker’s racing under motorsport, so the filter removed far more of some companies’ coverage than their peers’. From this edition no sector applies a topic filter. Relevance comes from the outlet allowlist, the word‑bounded name match, how early and how often the article names the company, and the rules above. Removing it leaves each sector’s mix of headline, lead and passing mentions within a few points of where it was.
A new baseline. These changes alter what each company is measured on, so this edition’s ranks are not compared with the previous edition’s. Movement is reported again from the next edition, when both are built the same way.
How it is scored
Every article runs through the same article-by-article engine as a client MRS®. Each article is tagged for sentiment (five levels, from GDELT tone), prominence (named in the headline, in the lead, or in passing) and salience (how many of the sector’s tracked companies it names), and scored on the four pillars: Sentiment 0.35, Distinction 0.25, Authority 0.25, Amplification 0.15. A company’s pillars are the average of its articles. Volume counts nowhere: a company with 5,937 stories and one with 46 are both judged on what their coverage says, where it ran and how central they are to it.
Sentiment is computed from the coverage. Amplification, Authority and Distinction are estimated from open data and carry a ~. Open data has no per-article reach, so every article inherits its outlet’s median reach and the reach percentile is a rank within the board’s own reach distribution, with tied outlets sharing the middle of their tie; it has no image or spokesperson tags, so Authority rests on the outlet’s credibility tier. Key Message alignment is not scored on the public board.
Reading the Sentiment pillar. A story that reads neutral scores 65 to 81 depending on the outlet it ran in, a positive one 68 to 96, a negative one 4 to 32. Across the board the pillar is almost exactly a straight line against the share of a company’s coverage that reads negative: about 5.1 points off for every tenth of its stories.
Weighted to priorities
Each reputation driver gets its own MRS®, scored on the articles tagged to it. The board ranks companies on those driver scores weighted to a set of illustrative priorities: Trust 30%, Responsibility 25%, Performance, Innovation and Leadership 15% each, the weighting of a company that puts reputation first. Your priorities will differ, and so will your ranking. That is the point of weighting.
Named in the headline or the lead
Every story is tagged for prominence already, so the board carries a second view of the same numbers: the score computed only on stories that name the company in the headline or inside the opening thousand characters. Switch it on above the board. It is a view, not the published rank, because a company does not get to disown a story for burying its name. But the distance between the two is itself a finding. A company that gains is carrying coverage it is barely in; one that loses is being helped by it.
How firm a rank is
Every sector is rescored 200 times on resamples of each brand’s own articles. Each brand card shows the range its sector rank falls in across those resamples, and the sector panel calls a lead clear only when the leader holds first place in at least 80% of them. Below that it says the sector is too close to call and names the rival. The rank itself is still printed: it is where the brand landed, not a claim that the gap is solid.
The overall column
Sector tone is not comparable: on this board the median Financials company scores higher than the median Health Care company, whatever either company does. Ranked on the raw score, this board puts three Financials names in the top seven. The overall order is therefore the distance between a brand’s score and the median score of its own sector, which is a question that survives the comparison: how far ahead of its own peer set is it. The score printed on every row is the unadjusted one, and sector ranks and the Gap are unchanged. Each sector’s norm is shown on the row and on the brand card.
The Reputation Gap
A brand’s coverage rank minus its tone rank, among its sector peers. It separates how much a brand is covered from how well that coverage reads. A brand is flagged when the distance is at least a quarter of its sector.
Who makes the board
At least 100 placements about the company and 4 sources about the brand in the window. A sector appears only with at least 8 brands, so a sector rank means something. The floor is never lowered to fill the board.
Reputation drivers
Trust, Innovation, Performance, Responsibility and Leadership are tagged from coverage themes. Each driver score is the full MRS® computed on the articles tagged to that driver. A driver is scored only on at least 12 distinct stories; below that it shows as n/a.
Why Trust reads lowest across the board
It does across the board, and it is not every sector sharing a trust problem. It is what gets tagged Trust. Litigation, regulation, recalls, investigations and breaches are negative events by definition, so the Trust bucket carries negative tone on any board in any month. Measured across every scored row on this board: A sector panel that named the lowest absolute driver would therefore print the same answer on every sector and tell you nothing, so each sector’s exposure is called on the distance between that sector and the full board on the same driver, which is the comparison that carries information. Where that gap is under half a point, or where it does not survive resampling the sector’s brands, the panel says so instead of naming a driver.
What this still gets wrong
Drivers are tagged from coverage themes, not read by a person, and the tagger cannot tell a company’s past from its present. Coverage of a former executive counts as the company’s, and where it is flattering it lifts the Leadership driver: 22 stories on this board, under 0.1% of the total, carry a headline describing someone as a former chief executive, chair or president. We have not stripped them, because the rules that would catch “Former PepsiCo CEO on her first job” also catch “Ford names former BlackRock executive”, which is current leadership news. The tagger also cannot tell a court win from a court loss: a company that defeats a claim and one that loses to it both read as litigation. Both are on the list for the tagged tier, and both are stated here rather than quietly absorbed.
A company reversing its own decision can read as good news, because the prose is about what happens next. On this board eleven headlines name a company walking back, scrapping or botching something, and two of them scored on the positive side. That is too few to justify a rule, so they are scored as they read and noted here.
Brands without drivers. A company whose coverage rarely carries a reputation driver (product and entertainment coverage often does not) is ranked on its plain MRS rather than the driver‑weighted score, and its card says so. This edition that applies to Nintendo.
Why some of the largest companies are not on this board. Every company here is published on the same rule: at least 100 placements in credible outlets that are verifiably about it. Among the largest US companies, this edition’s figures for those that fall short are: JPMorgan Chase (82 placements), Cigna (80), Dell (77), Elevance Health (76), Cisco (65), Merck & Co. (61), Citigroup (58), Bank of America (34), Walgreens (22), Humana (16), CVS Health (13), Target (11), Lowe's (6) and Procter & Gamble (held: open news data reaches too little of its coverage to rank it fairly). Over the 90 days the open news data the index reads returned no articles at all from Bloomberg, the Wall Street Journal, the Financial Times or the Associated Press’s own site, and five from Reuters: the outlets where much of the coverage of the largest banks and consumer companies runs. Outside the credible-outlet list, most of what does mention them is automated stock-filing pages. Every story here is also read, so a market page that mentions a bank or another company’s earnings call no longer counts for it. Publishing a company on a fraction of the evidence its peers are held to would make its rank meaningless, so the index does not. AT&T and 3M cannot be tracked at all: the open data does not tag their names. Search for any tracked company to see its figure.
The MRSIndex 100. The board is drawn from a named list of 100 companies in seven sectors adapted from the GICS groupings S&P and MSCI use: Health Care, Tech, Communications, Consumer, Financials, Automotive, and Industrials & Energy. A company is published only when at least 100 placements in credible outlets are about it over the 90 days. The list was reset for this edition, because reading every story changed what counts as coverage: 33 companies on the previous list no longer clear the floor once stories that are not about them are removed, among them JPMorgan Chase, Bank of America, Kraft Heinz and Mondelez. That says more about open news data than about them: it does not carry Reuters, Bloomberg or the Wall Street Journal, where much of their coverage runs. They move to a published reserve list and return when they clear. 33 companies join, most from the sectors added this edition. From now on, a company on the list that falls under the floor in an edition is replaced for that edition by the next reserve from its own sector, or from any sector if its own has none, and the swap is named here. This edition: Enbridge replaces Novartis (under the floor); Restaurant Brands International replaces AMD (under the floor); Tesco replaces Palantir (under the floor); Marks & Spencer replaces Best Buy (under the floor); Mattel replaces Nordstrom (under the floor); Lloyds Banking Group replaces Robinhood (under the floor).
Every story is read. Open data reads names, not meaning: a news feed matches a company wherever its name appears, so a market page that lists a stock, another company’s earnings, a crash report that names the car and a homepage that links to a story all arrive filed under the company. Rules catch the patterns. They do not catch the rest, so from this edition every story is read and judged: about the company, a substantive part of its own market, or not coverage of it. 149,104 story-company pairs were read for this edition, including every story behind every company on the board. Of the coverage that survived the rules above, 61% was judged not to be coverage of the company it had been matched to, and is not scored. The judgement is made from the headline, the web address and the outlet, because the open data publishes its coding of each article, not the article text, so a story is kept only where those show it is about the company. Stories are read by AI agents against a published review standard, with an independent second pass and a blind audit. The read decides which stories count; each counted story is then scored from a machine reading of the full article text. The full MRS® adds human analysts. A second, independent review judged 400 of them blind, drawn at random, and agreed on keep or drop 91% of the time. Every scored story is in the download so any one can be checked.
What open data does not carry
GDELT indexes the open web. Wire copy and subscription titles are thinly represented in it: Reuters, the Associated Press, Bloomberg, the Wall Street Journal, the Financial Times and the Economist together account for a negligible share of the coverage measured here, because syndicated wire stories are indexed under the outlet that ran them and paywalled sites are largely uncrawled. Any brand whose reputation is carried mainly in the wires and the paywalled press will read thinner here than it is.
Check the work
Every scored story is published with all of it: the outlet, its reach and credibility tier, the sentiment, prominence and salience codes, each of the four pillar sub‑scores, the article MRS® and the drivers it was tagged to. Download the scored‑article file (CSV) or the same data as a workbook, where every step of the engine is a live formula you can change. If you rebuild a number and get a different answer, we want to know.
Public Signal and the full MRS®
This page is the open, directional read. The full four‑pillar MRS®, human‑validated and priority‑weighted, runs on licensed and tagged coverage and will differ from the public signal. The framework, including the four pillars and the driver taxonomy, is set out in full in Reputation Intelligence (2026).
