# Method Producing script: `tools/report_caio_census_facts.py`, stages `search`, `combined`, `brand`, `shortlist`, `aggregate`. Working files: `data/raw/caio_serper/`, `data/raw/caio_serper_combined/` and `data/raw/caio_serper_brand/` (one JSON per company per pass, gitignored), `data/reports/caio_press_candidates.csv` (the shortlist the earlier passes read from) and `data/reports/caio_press_classified.csv` (the hand classification, 500 rows, with a `read_note` on every row). The aggregate stage refuses to run if any company is missing from the classified file. Snapshot date: 17 September 2026. Rebuilt 16 September 2026 after an adversarial review, and again on 17 September 2026 after a second review; every change from the first rebuild is in `changes_2026-09-16.csv` and every change from the second is in `changes_2026-09-17.csv`. ## Universe The 2026 Fortune 500 list, 500 companies with a domain each, from `installmap.com/data/universes/fortune500/companies.csv`. Every share in this dataset uses 500 as the denominator, because the census covers every company on the list. The LinkedIn columns come from a separate pass that covered 483 of the 500; where that pass is quoted the denominator is stated. Two ranks are shared. Fortune's own list gives Applied Materials and Hartford Insurance Group both rank 160 with no 161, and Community Health Systems and Huntington Bancshares both rank 350 with no 351, because each pair reports the same revenue. The rank column reproduces Fortune's list, so 500 companies carry 498 distinct ranks. This was checked against fortune.com's own ranking data on 17 September 2026. ## Searching Three passes, 1,016 Serper queries in total, 9,112 organic results stored. Results are cached per company per pass; deleting a cache file redoes it. **Pass 1, `caio`.** One query per company, `"" "chief AI officer"`, top 10 organic results kept. Where no organic result carried the phrase in its title or snippet, a second query, `"" "chief artificial intelligence officer"`, stored alongside. That happened for 4 companies, usaa.com, plains.com, kiewit.com and edwardjones.com, and all four of those second queries returned zero organic results, so that phrase put nothing into the census: five of the six phrases ran against all 500 companies and the sixth against four. A quoted query for the plain phrase returns spelled-out titles anyway, which is why Fidelity National Financial, Securian, Entergy and AGCO carry "Artificial Intelligence" in `title_as_stated` with `query_pass` = `caio`. 504 queries, 4,407 organic results. **Pass 2, `combined`.** One query per company for titles that fuse AI with another remit, which pass 1 structurally could not see: `"" ("chief data and AI officer" OR "chief AI and data officer" OR "chief digital and AI officer" OR "chief data, analytics and AI officer")`. 500 queries, 4,600 organic results. **Pass 3, `brand`.** Four companies returned zero organic results in pass 1 because the query used the Fortune list's legal name string: United Services Automobile Assn., Plains GP Holdings, Peter Kiewit Sons' and Jones Financial (Edward Jones). Each was re-queried under its common brand name (USAA, Plains All American Pipeline, Kiewit, Edward Jones) across all three phrase sets. 12 queries, 105 organic results. USAA and Edward Jones changed class; Plains and Kiewit stayed `none`. The `query_pass` column in `census.csv` says which pass produced the evidence on each classified row. ## Shortlisting, and why it was replaced by a reading pass Both query terms are quoted, so Google returns only pages carrying both, and almost every result contains the phrase somewhere. The phrase alone is therefore not a signal. The first two passes used a proximity test: a result was scored strong when it sat on the company's own domain, or when a distinctive token of the company name sat directly beside the phrase ("'s chief AI officer", "chief AI officer at ", " names ... chief AI officer"). In pass 1 that left 185 companies to read and batch-set the other 315 to `none` without anyone reading them. In pass 2 the same test, with the company-domain shortcut removed, left 102 companies to read. The test drops ordinary phrasings. "U.S. Bancorp's AI strategy is the most explicitly methodological of the large banks: a chief AI officer, Prashant Mehrotra, hired in 2025" scores weak, and U.S. Bancorp was batch-set to `none` with that sentence sitting in the shipped cache. **The reading pass, 17 September 2026.** Every company still in `none` or `unverified`, 437 of them, had all of its cached results read with no proximity filter: 7,912 organic results across the three passes, of which 5,422 carried a chief-level title string containing AI. Every company in the census, all 500, now carries a `read_note` in `census.csv` and in `data/reports/caio_press_classified.csv` recording what a person saw in its results. Companies classified in the earlier passes carry a note pointing at the url and quote columns that hold the result they were classified from. ## Classification rules - **appointed**: a company newsroom page, leadership page, annual report, SEC filing, wire press release, or a named news or trade outlet states the company has or appointed a Chief AI Officer (or Chief Artificial Intelligence Officer) with no unit, region or subsidiary in the title, and with no second remit fused into the title. Every appointed row carries a URL that was fetched and returned HTTP 200, with the sentence quoted in the `quote` column. - **combined**: the title carries AI alongside another remit and covers the enterprise: Chief Data and AI Officer, Chief Technology and AI Officer, Chief People and AI Enablement Officer. The exact title is recorded. The evidence has to be of the same grade as for appointed. - **A title outside the six searched phrases still counts** when a reader finds it on a source of the accepted grade. The reading pass did this for Lowe's, whose own executive leadership page states Executive Vice President, Chief Information and AI Officer; no query pass had returned that wording. - **divisional**: the same evidence, but the title names a unit, region or subsidiary. **This rule is applied to subsidiaries as well as internal units**, and it is applied everywhere. UnitedHealth Group is divisional because the title names Optum. Berkshire Hathaway is divisional because the title names TransRe, a Berkshire company. JPMorgan Chase is divisional because the fused title names J.P. Morgan Payments. Johnson & Johnson is divisional because the fused title names the global commercial organization, and Thermo Fisher Scientific because it names the Clinical Research Group. Marsh & McLennan is divisional because the role sits at Oliver Wyman, Equitable Holdings because it sits at AllianceBernstein, Honeywell because it sits at Honeywell Aerospace, and Salesforce because the title is scoped to Legal and Corporate Affairs. A divisional holder is never counted as an enterprise one. - **former**: the holder left or the role was folded. The class requires a source that states the title in the past tense, or a source that puts the same person in the same title at another company. A page showing the holder in post is not evidence of a departure. - **unverified**: the only claims come from contact-data vendors (ZoomInfo, RocketReach), analyst directories, honoree and award lists, conference or speaker bios, podcast listings, employee forums, listicles, vendor blogs or LinkedIn. These are recorded with what claims them, and are not counted as appointments. A company job posting that names no holder falls here too. - **none**: no result in any pass, read in full, places the title at the company. **The LinkedIn rule, one rule everywhere.** A title that lives only on a LinkedIn profile is `unverified` and is never counted as an appointment. That is how CVS Health, Labcorp and the other profile-only claims are handled, and since 17 September 2026 it is how Fastenal is handled too. Fastenal had been counted in the headline on a LinkedIn title alone, which contradicted the rule; it has left the enterprise count and the "or 54 counting Fastenal" variant of the widest number is gone. The enterprise count is press evidence only. The same rule covers a LinkedIn title that fuses AI with a second remit. Paccar and WESCO International were `none` until 17 September 2026 because the `linkedin_tier` column records only the rows whose title matched the plain Chief AI Officer regex, so their fused titles were invisible in `census.csv`. The LinkedIn pass holds Vikas Jain, "Chief Digital & Ai Officer", at Paccar and Sathish Srinivasan, "Vp- Chief Data, Ai Officer, Chief Cloud Officer (Cloud Coe) Api Edi Coe, Enterprise Architecture", at WESCO, both at tier `own_title_enterprise` with no press source of any kind. Both companies are now `unverified`, on the same rule as Fastenal. Their rows carry no URL, because a profile title is what the pass recorded and no page states it. **Published honoree lists are leads, not evidence.** CDO Magazine's Data and AI Leaders 100 and Constellation Research's Artificial Intelligence 150 were both read in full and matched against the universe. They are nomination-based rosters, and each disagrees with primary sources in checked cases: CDO Magazine's list gives PepsiCo's Magesh Bagavathi a chief title that PepsiCo and Bloomberg do not, and puts an AI title on a Starbucks role that CDO Magazine's own 2024 news story calls Chief Data and Analytics Officer; Constellation's list attributes the Department of War's chief digital and AI officer to Dow. Both were used to generate candidates, and every candidate was then classified on a primary source or left in `unverified`. Reported leadership-moves articles from the same outlets are ordinary trade-outlet evidence. Two titles held by one person (Global Chief Technology **and** Chief AI Officer at Dell, Chief Digital Officer **and** Chief Artificial Intelligence Officer at Fidelity National Financial) are treated as appointed, with the full title recorded. One compound title carrying two remits (Chief Data **and** AI Officer) is treated as combined. Where one company holds both, as Boeing does with a chief AI officer and a chief AI and data officer, it is counted once, in appointed. Where a company source and an outlet give different titles for the same person, the company source wins. That is what moved Expedia Group and Intuit from appointed to combined, and what keeps Humana and Yum Brands in `unverified`: Humana's own words are "Senior Vice President of Enterprise AI" and Yum's newsroom says "Chief Data Officer Cameron Davies", while the chief AI wording sits only on a profile or an award bio. The `title_as_stated` column holds the string the cited page carries, not a summary of it. Two rows that carried summaries were corrected on 17 September 2026: Berkshire Hathaway, where the TransRe page prints "Chief Artificial Intelligence Officer" with no company name, and Regeneron, where the cited Becker's page says "Regeneron's inaugural chief AI officer". Labcorp's title column is now blank, because the cited page is a job posting naming neither a person nor a title. ## Fetching Every page fetch went through the Webshare rotating gateway helper (`ai_sending_tool_v2/proxy_fetch.py`, `fetch_response_with_fallback`), with a browser user agent; no request came from the local IP. Serper is an API and was called directly. Every non-200 was retried twice and every status is recorded. All 24 appointed URLs and all 30 combined URLs were refetched on 17 September 2026. The statuses that were not a clean 200: - **Prudential Financial.** `news.prudential.com` sits behind Incapsula. Four attempts returned HTTP 200 each time, but one of the four served a 1,158-byte body with the quoted sentence absent. The row has CDO Magazine as a second source. - **LPL Financial Holdings.** The previously cited `investing.com` URL returned 500, 500 and then 200 on 17 September, and returned 500 twice and 403 once in the review. The row now cites LPL's own press release, fetched at 200. - **Masco.** The Business Wire canonical URL returned 500 on three attempts, so the row cites a syndicated copy of the same release, fetched at 200. - **Labcorp Holdings.** ZoomInfo and a job-board mirror each returned 500 on three attempts, so the strings they carry could not be read on a page. - **LinkedIn profiles** return 200 behind a sign-in wall. Where a row rests on one (Coca-Cola, CDW, APi Group, Procter & Gamble), the note says the string comes from the cached search result rather than from the page. - **Reddit and Facebook** served bot interstitials or login walls for Resideo and Voya, and both rows say so. - **The Becker's page** behind the Regeneron row returned 200, then 500, then 200 again over three attempts on 17 September 2026. - **RTX.** The newsroom URL returned a 7,473-byte body with a 500 on the first attempt and 63,438 bytes at 200 on the retry, with the quoted sentence present. - **PepsiCo's previous source**, a Getty Images caption page, returned 200 with a 26 KB body whose visible text is a bot-validation notice, so the caption could not be read. The row moved to `unverified` on a Bloomberg page that gives a different, non-chief title. - One repeater, a Facebook post, returned 200 with an empty body behind a login wall; `repeaters.csv` records that its sentence comes from a search snippet. The fetch counts a reader can check against the shipped files are the cited URLs: `changes_2026-09-16.csv` holds 37 distinct URLs and `changes_2026-09-17.csv` holds 51, every one of them fetched through the gateway. In the 16 September rebuild two URLs returned 500 twice (businesswire.com's TransRe release and an Economist podcast page) and both were replaced. ## Merging with the LinkedIn pass `linkedin_tier` comes from `data/reports/caio_matches_reviewed.csv`, the hand-inspected pass over vendor people rows joined on company domain, restricted to rows whose title matched the Chief AI Officer regex and survived inspection. Eleven company domains carry such a match. `has_caio_enterprise` is 1 when and only when the press census says `appointed`. The `evidence` column reads `press`, `both` (press plus a LinkedIn enterprise title), or blank. A LinkedIn enterprise title with no appointment sits in `linkedin_tier` and is visible in the file, but it adds nothing to the count: UnitedHealth Group, Entergy and Fastenal are the three rows in that position. All 381 `none` companies have no Chief AI Officer title match in the LinkedIn pass. 368 of the 381 were covered by that pass, which holds person rows for 483 of the 500; for the other 13 there is no LinkedIn data either way. The two fused-title profiles the pass holds, at Paccar and WESCO International, moved those companies out of `none` on 17 September 2026; the 13 with no coverage are unchanged. ## Limitations - **Six phrases, not every title, and one of the six barely ran.** The three passes searched "chief AI officer", "chief data and AI officer", "chief AI and data officer", "chief digital and AI officer" and "chief data, analytics and AI officer" against all 500 companies, and "chief artificial intelligence officer" against the 4 whose first query returned no result carrying the plain phrase, all 4 of which came back empty. Any title outside those phrases is under-counted or missed. ADP's "Chief Product and AI Officer" and Lumen's "Chief People and AI Enablement Officer" are in this file only because the plain-phrase query happened to return them anyway, and Lowe's "Chief Information and AI Officer" is in it only because a reader opened the company's leadership page. The count is a floor. - Serper returns the top 10 organic results per query. An appointment covered only below the tenth result is missed. The reading pass reads the results the query returned, not the ones it did not. - An appointment can be real and unannounced. `none` means no public record was found in three passes and none was found when a person read the cached results. - 39 companies sit in `unverified`, where the title is claimed only on a vendor page, directory, conference bio, award bio or LinkedIn. If half of those are real, the enterprise count rises. 36 of the 39 carry the page that claims them; Fastenal, Paccar and WESCO International rest on a LinkedIn profile title with no page. - Press coverage skews to large and newsworthy companies. A quiet appointment at a mid-list industrial or utility is less likely to be written about than the same appointment at a retailer. - LinkedIn titles are self-reported and undated. The pass they come from cannot show that a holder has left, which is what happened at UnitedHealth Group and Entergy, and it cannot show a unit the profile headline omits. - Published honoree lists are leads, not evidence, for the reasons set out above. - Titles are unstable. Twelve companies in this list changed the title, folded it into a data role, or lost the holder, and one appointed row, Masco, dates from two days before the snapshot. - Dates are the date of the announcement or of the article that reported it, whichever the source gives. Where only an article date or a JSON-LD date exists it is marked. Three appointed rows carry no date because the page stating the title gives none. - Some sources are trade outlets, company leadership pages or partner releases rather than national news. The `source_type` column says which, so any stricter count can be recomputed from the file. One appointed row, Applied Materials, rests on a paid Business Wire release issued by a partner rather than by the company, and one, Securian, rests on a company annual report where the holder's rank is Second Vice President. ## Revisions This section holds the revision history that used to sit in the body of the [research post](https://professionlens.com/research/fortune-500-chief-ai-officer-census). The post was shortened on 23 September 2026; no number changed. ### What the first version of the search missed The first objection to a census like this is that the search missed people, and in an earlier version it did. We searched five exact title phrases against all 500 companies, plus a sixth, "chief artificial intelligence officer", against four: "chief AI officer" in the first query pass, then "chief data and AI officer", "chief AI and data officer", "chief digital and AI officer" and "chief data, analytics and AI officer" in a second pass. The spelled-out phrase ran only where the first query returned no organic result carrying the plain one, which happened at four companies, and all four of those queries came back empty, so it put nothing into the census. A quoted query for the plain phrase returns the spelled-out wording anyway, which is how Fidelity National Financial, Securian, Entergy and AGCO were found: each carries "Artificial Intelligence" in the title and each was found on the plain phrase. A third pass re-queried the four companies whose legal name string on the Fortune list returned nothing, under the names a person would type: USAA, Plains All American Pipeline, Kiewit and Edward Jones. Then we shortlisted which companies a person would read, using a proximity test that kept a result only when it sat on the company's own domain or when a distinctive company token sat directly beside the phrase. In pass 1 that left 185 companies to read and batch-set the other 315 to none with nobody reading them. A reviewer sampled 20 of those 315 and found four companies that hold or held the title, with the evidence sitting in the cache we had already shipped. The first cached result for U.S. Bancorp reads "U.S. Bancorp's AI strategy is the most explicitly methodological of the large banks: a chief AI officer, Prashant Mehrotra, hired in 2025 from ...", which the proximity test scores weak. So we threw the test away. Every company still in none or unverified, 437 of them, had all of its cached results read with no filter at all: 7,912 organic results, of which 5,422 carried a chief-level title string containing AI. Every one of the 500 companies now carries a `read_note` column in census.csv saying what its results contained. Two blind spots are left. We cannot see an appointment nobody announced, and Serper returns the top ten organic results per query, so coverage that sits below the tenth result is missed whether or not a person reads the page. ### Class moves in the 17 September pass Nine of the 30 rows are new since the 16 September version. The reading pass moved six into this class: Dominion Energy, Lowe's and Masco from none, RTX, Equinix and Lumen Technologies from unverified. A cross-check against CDO Magazine's published honoree list moved three more: Cigna Group, Hartford Insurance Group and BNY. Fastenal and PepsiCo moved into this class from opposite directions. Fastenal, rank 475, has a Chief Artificial Intelligence Officer in the LinkedIn pass and nothing in any of the three query passes, so it sits in unverified rather than in the headline, which is where CVS Health and Labcorp Holdings already sat under the same rule. The 16 September version counted Fastenal in the headline on that LinkedIn title, and a reviewer pointed out that the same evidence was being read two ways. PepsiCo, rank 46, moved into unverified from the other direction: its row used to cite a Getty Images caption whose page returns a bot-validation body, and the best readable source, a Bloomberg video page, gives Magesh Bagavathi the title Senior Vice President and Global Head of Data, Analytics and AI, which is not a chief-level AI title. 36 of the 39 rows in census.csv carry the page that claims them, so a reader can work through those. The other three, Fastenal, Paccar and WESCO International, rest on a title the LinkedIn pass recorded with no URL beside it. Paccar and WESCO moved here in this pass: the LinkedIn pass gives Paccar's Vikas Jain "Chief Digital & Ai Officer" and WESCO's Sathish Srinivasan a title beginning "Vp- Chief Data, Ai Officer", both scored enterprise-wide, and no press source of any kind carries either. Every other source a reviewer checked gives Jain Chief Digital Officer and Srinivasan Vice President and CIO, with no AI in the title. ### Three reviews and what they changed Three adversarial reviews have been run against this file. The first recomputed the census, refetched every classified row and queried 17 none companies, which is how Prudential Financial and IBM came in; the Prudential miss pointed at the method rather than at one row, since two query phrases cannot see a Chief Data and AI Officer, and that is why the four fused phrases were run against all 500 companies. changes_2026-09-16.csv records 39 changes from that rebuild, 30 of which moved a company's press status. The second review recomputed every number again, refetched 47 URLs and ran 20 fresh queries against companies we had marked none. Four of the 20 held or had held the title, and in all four cases the evidence was in the pass-1 cache we had already shipped, dropped by the proximity test. U.S. Bancorp is now an appointed row and changed the headline. ADP is now a former row. Carrier Global is now unverified, on a conference roster and a CDO Magazine list rather than a company source. Starbucks stays none: Das Dasgupta is in the combined table at McDonald's, and CDO Magazine's own 2024 story about his previous job is headed "Starbucks Appoints Das Dasgupta, PhD as Chief Data and Analytics Officer", without AI in the title. The third review recomputed every number a third time, refetched 27 URLs, ran 25 fresh queries against companies marked none and found no missed appointment. It moved Paccar and WESCO International out of none, corrected the count of titles the fused phrases added, and corrected the sentences about which pages repeat which figure. changes_2026-09-17.csv records 61 changes with the old value, the new value and the reason. 51 of the 61 carry a URL; the other ten record a company that stays where it is, or a move made on the LinkedIn pass, where there is no page to cite. 36 of the changes moved a company's press status: - The reading pass moved 27 companies. Three to appointed: U.S. Bancorp and Securian Financial Group from none, Apollo Global Management from unverified. Six to combined: Dominion Energy, Lowe's and Masco from none, RTX, Equinix and Lumen Technologies from unverified. Three to divisional: Salesforce, Marsh & McLennan and Equitable Holdings. Two to former: Automatic Data Processing and Western Digital. And 13 from none to unverified, where a named person carries an AI chief title on a directory, a conference bio, an award bio, a podcast listing, a forum post or LinkedIn: AES, APi Group, Amazon, CDW, Carrier Global, Coca-Cola, Fidelity National Information (FIS), KLA, Northern Trust, Northrop Grumman, Reinsurance Group of America, Resideo Technologies and Voya Financial. - The CDO Magazine cross-check moved five more: Cigna Group, Hartford Insurance Group and BNY to combined, Honeywell to divisional, Procter & Gamble to unverified. - Four moves came from reviews. The second review moved Fastenal from none to unverified under the LinkedIn rule and PepsiCo from combined to unverified. The third review moved Paccar and WESCO International from none to unverified, because the LinkedIn pass holds a fused enterprise chief AI title for each and no press source of any kind carries it, which is Fastenal's position exactly. The other 25 changes left the class alone. They swapped an unreadable source for a readable one, corrected a title string to what the cited page actually says, as at Berkshire Hathaway and Regeneron where the column had carried a summary, or recorded what the reading pass found at a company that stays none. Three of those are worth a line each. At Morgan Stanley, Jeff McMillan was promoted in March 2024 into a newly created firmwide AI role, and podcast and event pages call him the firm's Chief AI Officer, but the internal memo CNBC reproduced gives the title as "Head of Firmwide Artificial Intelligence", and a Brillio release of 2 June 2026 calls him "former Head of Firmwide AI at Morgan Stanley", so the company stays none. At First Citizens BancShares, the American Banker sentence naming "Kristin Milchanowski, the bank's chief AI and data officer" beside First Citizens is about BMO, which sold 138 US branches to First Citizens. Humana and Yum Brands stay in unverified with corrected title strings: Humana's own words are "Senior Vice President of Enterprise AI" and Yum's own newsroom says "Chief Data Officer Cameron Davies", and in both cases the chief AI wording exists only on a profile or an award bio. We also checked the census against three published lists, and we treat those lists as leads rather than as evidence. CDO Magazine's Data and AI Leaders 100 for North America 2026 carries 72 entries at 71 Fortune 500 companies, 27 of them giving a chief-level title containing AI at 26 companies. 18 of the 26 were already classified as something other than none, and 8 were none. Checking each of the 8 against a primary source moved 5 into a class with press evidence and 2 into unverified, and left AIG at none, because its entry reads "Global Chief Data Officer & Head, Agentic AI Solutions" and the chief title itself carries no AI. The same list disagrees with primary sources twice: it gives PepsiCo's Magesh Bagavathi "SVP, Global Head of Data, Analytics & AI", not the Chief Data and AI Officer title we had been citing from a Getty caption, and its 2026 list puts Das Dasgupta at McDonald's with an AI title while its own 2024 story puts him at Starbucks as Chief Data and Analytics Officer. CDO Magazine's reported leadership-moves articles are a different thing and are treated as trade-outlet evidence, as before. Constellation Research's Artificial Intelligence 150 2026-2027 names 34 Fortune 500 companies, 4 with a chief-level AI title: Boeing and Leidos Holdings, both already appointed; Nike, already unverified, where Constellation names a different person from the one our Nike row names, with the same title; and Dow, where the person it lists, Cameron Stanley, is the US Department of War's chief digital and AI officer in every cached result, so Dow stays none. The Constellation list produced no name the census did not already hold. AI Magazine's Top 10 Chief AI Officers, in two editions, adds no Fortune 500 name we did not already hold, and its 2025 edition headlines Alan John as "CAIO of Nike" while its own body calls him Global Head of Data and AI at Nike. Not every fetch succeeds, and the file records the ones that did not. Prudential Financial's newsroom sits behind Incapsula: four attempts on 17 September returned 200 each time, but one of the four served a 1,158-byte body with the quoted sentence absent. LPL Financial's previously cited investing.com URL returned 500, 500 and then 200 on 17 September, and 500 twice and 403 once in the review, so the row now cites LPL's own press release, which returned 200. Masco's Business Wire canonical URL returned 500 on three attempts, so the row cites a syndicated copy of the same release. The Becker's page behind the Regeneron row returned 200, then 500, then 200 again on 17 September, and RTX's newsroom URL returned a 7,473-byte 500 body on the first attempt and 63,438 bytes at 200 on the retry, with the quoted sentence present. Labcorp's ZoomInfo page and a job-board mirror each returned 500 on three attempts, so its title column is blank and the note explains why. LinkedIn profile pages return 200 behind a sign-in wall, and where a row rests on one the note says the string comes from the cached search result rather than from the page, at Coca-Cola, CDW and APi Group. Reddit and Facebook returned bot interstitials or login walls for Resideo and Voya, and those rows say so. ### Full funnel | Step | Number | |---|---| | Companies in the universe | 500 | | Companies classified by hand | 500 | | Companies whose results were read by a person | 500 of 500 | | Serper queries, pass 1 | 504 | | Organic results, pass 1 | 4,407 | | Serper queries, pass 2 | 500 | | Organic results, pass 2 | 4,600 | | Serper queries, pass 3 | 12 | | Organic results, pass 3 | 105 | | Serper queries in total | 1,016 | | Organic results in total | 9,112 | | Companies read individually in pass 1, before the reading pass | 185 | | Companies batch-set to none in pass 1, before the reading pass | 315 | | Companies re-read with no filter on 17 September | 437 | | Organic results read in that pass | 7,912 | | Of those, carrying a chief-level title string containing AI | 5,422 | | Companies with zero organic results on the legal-name query | 4 | | Distinct URLs cited in changes_2026-09-16.csv | 37 | | Distinct URLs cited in changes_2026-09-17.csv | 51 | | Field changes recorded on 16 September | 39 | | Field changes recorded on 17 September | 61 | Four companies returned zero organic results in pass 1, because the query used the Fortune list's name string: United Services Automobile Assn., Plains GP Holdings, Peter Kiewit Sons' and Jones Financial (Edward Jones). Four returned zero in pass 2: Performance Food Group, Lithia Motors, Jones Financial and Interactive Brokers Group. The brand re-query found USAA a combined title and Edward Jones a former one; Plains and Kiewit stay none. ## Licence CC BY 4.0. Attribution: ProfessionLens.