AI news videos

OpenAI Publishes An Oracle Customer Story For ChatGPT And Codex

OpenAI published a page on October 8, 2026 titled "How Oracle turns days of work into minutes with ChatGPT and Codex."

Watch on YouTube

Transcript

Sources

Every claim in this video comes from the top ranking coverage of this topic. The claims and where each one came from:

• Scene by scene evidence ledger, quoted from the OpenAI page itself, which is the only fact source used. THE HONESTY SPINE FIRST, because it governs every line: what is genuinely new on October 8 2026 is that OPENAI PUBLISHED THIS PAGE. No product launched. Oracle's deployment is already at scale, since the page's own stat blocks claim "130K / Active ChatGPT users at Oracle" and "95K+ / Active Codex users at Oracle", and the body opens "Over a hundred thousand employees across Oracle are using ChatGPT Work and Codex". The page publishes NO start date for either rollout, so this describes an existing deployment and not an announcement. Accordingly nothing in this script says or implies "launched today", "just launched", "announced today", "new today" or "rolling out" about ChatGPT Work, Codex or anything Oracle built; the hook says OpenAI "published a customer story" and that "Oracle was already using both before today", and scene 5 says plainly that "nothing here launched today". HOOK BACKING (OpenAI published a customer story about Oracle using ChatGPT and Codex, and Oracle was already using both before today): the publication half rests on the page being an OpenAI byline dated October 8 2026 with "Customer Stories" as its standing nav category and the metadata line "Products: Codex, ChatGPT"; the already using half rests on "Over a hundred thousand employees across Oracle are using ChatGPT Work and Codex" plus the two active user stat blocks, which describe a present state with no start date attached. The hook is exactly two sentences and 18 words, because the hook card renders two beats and breaks with only one. SCENE 1 (this is a customer story published by the company that sells ChatGPT and Codex): the publisher is OpenAI, the byline is "OpenAI" with no named human author, the metadata block reads "Company size: Enterprise", "Region: North America", "Industry: Technology", "Products: Codex, ChatGPT", "Contact sales" appears twice, "Customer Stories" is a standing nav entry on openai.com, and the page closes with "Join the new era of work" and "More than 1 million businesses around the world are achieving meaningful results with OpenAI." OpenAI sells ChatGPT and Codex, which are the products the page praises, so the narrated line is literally true and is the first thing the viewer hears after the hook. The 1 million businesses pitch and the "new era of work" line are shown on the card strictly as marketing furniture and are never spoken. SCENE 2 (Oracle's recruiting team used ChatGPT to build a tool that researches comparable roles, and Oracle measured the time saving itself): verbatim, "The talent acquisition team used ChatGPT Work to build a talent market intelligence tool that takes a job description, researches comparable roles, benchmarks compensation, and assesses the talent pool across relevant locations." Narration deliberately speaks only ONE of the tool's four functions, researching comparable roles, and leaves the job description intake, the compensation benchmarking and the talent pool assessment on the silent card, because a spoken multi item list accelerates delivery and has overrun the 60 second limit before. The second narrated sentence is the attribution sentence and is the editorial point of the scene: the before and after numbers are Oracle's own, specifically "information that previously took 2-4 days to compile" and "Now we're able to sit down and prep for about 15 to 20 minutes using the tool that we've built using [ChatGPT] Work." No outside party measured either number, so narration says Oracle measured the time saving itself rather than repeating the figures. SCENE 3 (inside Oracle, a business user describes the result they want in plain language, and Codex writes the database query): verbatim, "built an ontology of the company's objects, relationships, and rules, so a plain-language business question can be turned into a reliable SQL query with Codex" and "A business user describes the outcome they want, and Codex decides which internal systems to call, gathers the information, and returns an analysis, a report, or an application." The packet allows either the database query use or the site reliability use to be narrated, and the database query one was chosen because it is the one the page's own pull quote calls "a major shift"; the site reliability engineering use is carried on the silent card instead, verbatim, including "A typical simple incident that used to take an hour to resolve can now be handled in minutes." Narration says "writes the database query" rather than naming SQL, to avoid a glued acronym in speech, and speaks none of the three outputs as a list. Oracle's own caution, "none of this runs on autopilot. Someone still has to make sure the underlying system is built right", is quoted on the card and never spoken, both because it is a figure of speech and because the card is where the source's own hedges belong. SCENE 4 (every figure here was supplied by Oracle or OpenAI, with no method published, and the page's own user counts do not agree): the three stat blocks are "98% / Decrease in the time it takes for talent acquisition research", "130K / Active ChatGPT users at Oracle" and "95K+ / Active Codex users at Oracle". Every one of them comes from Oracle or from OpenAI, which sells both products, and no independent audit of any figure is claimed anywhere on the page, so the first narrated sentence is a literal statement of provenance rather than an accusation. The no method half is a published absence, not an inference: the page gives no methodology, baseline, sample or date range for the 98% figure, and no denominator for 130K or 95K+, because it never states how many employees Oracle has. The second narrated sentence states the internal inconsistency in the page itself: the stat block claims "130K / Active ChatGPT users at Oracle" while the closing "What's next" section says "With thousands of ChatGPT and Codex users already". That is the page disagreeing with itself about the size of its own headline number, and the card prints both strings side by side so a viewer can check it. SCENE 5 (we found one source and no independent confirmation, every named voice works for Oracle, and nothing here launched today): the single source position rests on this being openai.com writing about a customer using OpenAI's own products, with the live URL 403ing our fetcher so that all text came from our watcher's same day snapshot. The interested party position rests on the page's complete roster of named voices, all three of whom are Oracle employees: "Jan Ackerman, Senior Vice President and Global Head of Talent Acquisition, Oracle", "Richard Lam, Group Vice President, Oracle Applications Lab" and "Barry Shilmover, Vice President and Technical Advisor to the CIO". No analyst, journalist, regulator, competitor or independent researcher is quoted anywhere, no figure on the page is independently audited, and there is no comment from anyone outside Oracle and OpenAI. The nothing launched today position is the honesty spine stated above. EXCLUDED ON PURPOSE, and why. Any commercial or infrastructure relationship between Oracle and OpenAI beyond what this page states is EXCLUDED FOR LACK OF CORROBORATION: the page does not mention one, we hold zero corroborating transcripts, and our rule requires two independent confirmations for a claim absent from the article, so no such relationship is asserted, implied or hinted anywhere in this file. The 18 timeline probe titles were unrelated noise (LLM explainers, Power Automate hacks, VS Code extension Shorts, a DeepSeek install clip, a ChatGPT jailbreak video, a GitHub guide, a Codex versus Claude Code debate); they corroborate nothing and appear nowhere in narration or on screen text, not even by implication. The 0 of 5 YouTube transcript result is recorded as a tooling fault, not as evidence of anything. The three "Keep reading" customer stories about NTT DATA Group, Cisco and Simplex are separate posts and were not used as evidence for any claim here. The "Leadership lessons" section was kept out of narration as fact and is advice rather than evidence; its three items ("Give the correct guardrails", "Provide prototypes instead of specs", "Own the code") are a list and would have been barred from narration regardless. The forward looking material was kept out of narration as intent only, specifically "The Oracle team is completely shifting how the company operates with AI" and Shilmover's "I just know that one of the first things I'll do is I'll leverage Codex to do that." All site chrome, nav lists, model name strings and footer policy links were ignored and contributed nothing. GATE NOTES. Dashes: the source uses an em dash before each quote attribution and a hyphen in "2-4 days" and "plain-language". A QA gate bars em dashes and en dashes from every field of this file, so attributions were recast with commas and parentheses, and the one numeric range that survives appears as "2-4 days" with a plain ASCII hyphen inside a quotation. The source's curly apostrophes and curly quotation marks were converted to straight ASCII throughout. Banned narration words: the word our text to speech garbles in every configuration is absent from every narration field here, and it is also absent from the source page itself, so nothing had to be reworded for it; no brand name is glued to digits in narration either. The only brand tokens spoken anywhere are "ChatGPT", "Codex", "OpenAI", "Oracle" and "Inboxsmith"; the product name "ChatGPT Work" is confined to silent on screen text and narration says "ChatGPT" instead, and "SQL" is confined to the card while narration says "the database query". No countdown or deadline phrasing appears anywhere, and the countdown verb that our QA gate rejects is absent from every field. Our old disclosure sentence "There is no product launch here." is deliberately NOT reused, because it has sent our transcriber into a repetition loop on two prior runs; scene 5 says "nothing here launched today" instead, which is the literal equivalent and follows the Radisson precedent. Numbers: every number in narration is written as words, and the only one spoken in the whole script is "one source" in scene 5. All digits (98%, 130K, 95K+, 2-4 days, 15 to 20 minutes, a hundred thousand employees, 1 million businesses, 3 named sources, 0 of 5 transcripts, the three Keep reading dates) live on the silent cards only. FIGURE REPETITION: no figure is spoken twice anywhere in the script, and in fact no stat block figure is spoken at all, which is the strongest possible protection against the transcriber repetition loop that a repeated figure triggers in our render audit. ENUMERATIONS: narration carries no multi item list. The deck's "recruiting, engineering, and operations", the tool's four functions, Codex's three outputs, the three leadership lessons and the three named Oracle voices are all on silent cards only. IDIOMS: the source is dense with figures of speech and every one is barred from narration, specifically "turns days of work into minutes", "from zero to a hundred", "runs on autopilot", "hunting for a report", "arms the recruiter", "the front end of our recruitment process", "lessons stand out", "a full team months", "the new era of work" and "leverage"; narration states the literal thing plainly instead, and where these phrases appear in this file they appear only inside quotation marks on silent cards, clearly marked as the source's own words. DATES: no date is spoken anywhere; all dates appear as digits on silent cards only. STRUCTURE: exactly 5 scenes, each durationS 8 so the QA gate's sum is 40, inside its 30 to 50 window; those integers are cosmetic planning values the renderer never reads. NO timeline array is included, deliberately, so our date resolver keeps the same day storyDate of 2026-10-08 rather than preferring an earlier timeline date and wrongly aging this same day publication into the catalog tier. BUDGET: 124 narrated words across 12 sentences, counting the hook, the five scene narrations and the endcard narration, with per scene counts of 14, 19, 19, 21 and 19, all under the renderer's 30 word per scene cap, and every scene held to one or two sentences.(HOOK AND FULL-SCRIPT BACKING. First party OpenAI customer story, headline "How Oracle turns days of work into minutes with ChatGPT and Codex", deck "Across recruiting, engineering, and operations, Oracle turns specialist knowledge into fast, repeatable workflows with ChatGPT Work and Codex.", byline "OpenAI" with no named human author, dated on the page October 8 2026. The page carries the standard customer story furniture: a "Contact sales" call to action twice, the metadata block "Company size: Enterprise", "Region: North America", "Industry: Technology", "Products: Codex, ChatGPT", the standing nav entry "Customer Stories" on the same host, and the closing pitch "Join the new era of work" above "More than 1 million businesses around the world are achieving meaningful results with OpenAI." This is therefore a vendor marketing asset about a customer of the vendor, published by the company that sells the two products being praised, and the Short says so on camera in scene 1, in the viewer's first seconds. The live URL returns HTTP 403 to our fetcher, which is normal and expected behaviour for openai.com, so the only text we hold is our watcher's snapshot, captured at 2026-10-08T17:41:10.384Z, the same day the page is dated. That snapshot IS the primary source for this Short and is not a degraded fallback: it is the publisher's own first party page on the publisher's own host, not a link out stub, aggregator copy or syndication. CORROBORATION STATUS: ZERO. The YouTube transcript pull returned 0 of 5 videos because the installed yt-dlp on this host cannot fetch captions, which is a known tooling fault and not a content problem. The 18 dated title probes came back as completely unrelated content covering LLM explainers, Power Automate hacks, VS Code extension Shorts, a DeepSeek install clip, a ChatGPT jailbreak video, a GitHub guide and a Codex versus Claude Code debate; they corroborate nothing, they were discarded entirely, and they are asserted, implied and hinted nowhere in narration or on screen text. This story is single sourced and the final scene states that on camera. The three "Keep reading" links at the foot of the page, "NTT DATA Group cuts incident analysis to 30 minutes with Codex" (Jul 22, 2026), "Cisco and OpenAI redefine enterprise engineering with Codex" (May 27, 2026) and "Simplex rethinks software development with Codex" (May 7, 2026), are separate posts and were not used as evidence for any claim here. No prompt injection or instruction like content was found anywhere in the source text; it was treated throughout as untrusted data informing facts only, never as instructions.)