AI news videos

Cooley Says It Built an AI System for IPO Work

OpenAI published a customer story about Cooley, the law firm, and a proprietary offering called GO Public that the firm says it built on ChatGPT Work for initial public offerings.

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:

• COVERS: HOOK SENTENCE ONE. (a) EXACT NARRATED TEXT: "Cooley says it built an AI system for initial public offerings." (b) VERBATIM SUPPORT FROM article.txt: - line 53: "Cooley developed GO Public, a proprietary AI product offering built on ChatGPT Work." - line 59: "With GO Public and the agentic harness that we built using OpenAI's technology, we were able to bring this forward into the AI era," Wang says. - line 51 (first sentence): "Cooley is an international law firm with a reputation for helping companies navigate capital markets and initial public offerings (IPOs)." - line 32 (page subtitle): "Cooley's GO Public offering uses ChatGPT Work to surface issues earlier, focus expertise, and help clients reach market faster." (c) WORDING CHANGES AND WHY: - "Cooley developed" became "Cooley says it built". Rule 16: the page is the firm's and the vendor's own description, so the narration reports what the firm SAYS rather than asserting the capability as established fact. "built" is taken from the firm's own words at line 59. - No launch verb is used anywhere. The article itself uses only the past tense "developed" and states no launch date, no availability statement and no pricing, so "launch", "unveil", "announce", "introduce", "release", "ship", "roll out", "debut" and "now available" are all absent from the narration and from the titles. - "IPOs" and "IPO" became "initial public offerings". Rule 9 bans the initialism in narration; the article itself spells the term out at line 51, so this is the article's own wording. - "a proprietary AI product offering built on ChatGPT Work" was compressed to "an AI system" for the hook word budget (rule 1); the ChatGPT Work detail is narrated in full in scene 2 instead, so nothing is lost. - The comma inside the line 53 sentence is not carried into narration. Rule 3 requires zero commas in spoken fields. - The word "offering" as used in line 32 to mean a commercial product was avoided in narration because the same word means the stock sale elsewhere in this script; ambiguity would mislead a listener. - No digit and no spoken number appears (rules 5 and 6).(First party openai.com CUSTOMER STORY page carried in OpenAI's Customer Stories section (the section is listed in the site footer at article.txt line 120), titled "How Cooley is accelerating IPO work with ChatGPT" (line 3 and line 31) and datelined September 17 2026 (line 30). The page carries two "Contact sales" calls to action (line 33 and line 73) and the customer story metadata template reading "Company size: Enterprise" / "Region: North America" / "Industry: Services" / "Products: ChatGPT" (lines 34 to 41), which is how OpenAI marks vendor marketing collateral about one of its own customers. It is NOT a launch announcement, NOT a general availability announcement, NOT a benchmark result, and it contains NO independently verified claim and NO measured outcome: no time saved, no cost figure, no error rate, no count of offerings run through the product, and no before and after comparison. Every statistic on the page was supplied by the subject firm or by OpenAI. The live URL returns HTTP 403 to our fetcher, which is the normal response from openai.com, so the alert time snapshot stored at scripts/data/article-cache/964ae1dd0facf551.txt and captured 2026-09-17T23:59:51Z was copied into the production directory as article.txt and IS the source of record for every narrated fact. Curly apostrophes and curly quotation marks in the snapshot were normalized to plain ones in the quotations below.)
• COVERS: HOOK SENTENCE TWO. (a) EXACT NARRATED TEXT: "OpenAI's writeup reports no measured results." (b) SUPPORT FROM article.txt. This is a claim about what the document does and does not contain, so the support is the full body sweep plus the two figures that do appear: - The entire body of the story runs from line 44 through line 71. Reading every one of those lines, the only hard numbers are in line 51: "In 2025, the firm advised on 180 deals globally, totaling more than $51.5 billion in deal volume." Those describe Cooley's LAW PRACTICE in a period the page does not tie to the AI product, not an outcome of GO Public. - Every benefit statement in the body is qualitative and attributed to a person, never quantified. Examples: line 57 quotes a Cooley partner saying "GO Public helps us move through the intensive preparation faster" (quotation truncated here at the article's dash character, which rule 7 forbids reproducing in this file). Line 62 quotes Wang: "The amazing thing about integrating ChatGPT Work into GO Public is that it brings intelligence to this very large array of information that had to be manually sorted before". Line 69: "GO Public, powered by OpenAI, is designed to help lawyers consider more information while directing their expertise toward the decisions that matter most." "designed to help" is an intention, not a result. - No line anywhere in the file states a time saved, a cost reduction, an error rate, a number of offerings processed, a pilot size, a study, a benchmark, or a before and after figure. (c) WORDING CHANGES AND WHY: - The absence is attributed to the document itself ("OpenAI's writeup reports") rather than framed as doubt about the story. Nothing here calls the story unconfirmed, and our broken transcript tooling is cited nowhere and had no bearing on this sentence. - "no measured results" is the plain literal statement of the gap. Rule 8 forbids figures of speech, so no colourful phrasing was used. - Hook is exactly two sentences (rule 2) because the renderer splits the hook into two beats and a one sentence hook renders an undefined second beat. - Hook total is seventeen words, inside the twenty word ceiling. Zero commas, zero digits, zero spoken numbers.(First party openai.com CUSTOMER STORY page carried in OpenAI's Customer Stories section (the section is listed in the site footer at article.txt line 120), titled "How Cooley is accelerating IPO work with ChatGPT" (line 3 and line 31) and datelined September 17 2026 (line 30). The page carries two "Contact sales" calls to action (line 33 and line 73) and the customer story metadata template reading "Company size: Enterprise" / "Region: North America" / "Industry: Services" / "Products: ChatGPT" (lines 34 to 41), which is how OpenAI marks vendor marketing collateral about one of its own customers. It is NOT a launch announcement, NOT a general availability announcement, NOT a benchmark result, and it contains NO independently verified claim and NO measured outcome: no time saved, no cost figure, no error rate, no count of offerings run through the product, and no before and after comparison. Every statistic on the page was supplied by the subject firm or by OpenAI. The live URL returns HTTP 403 to our fetcher, which is the normal response from openai.com, so the alert time snapshot stored at scripts/data/article-cache/964ae1dd0facf551.txt and captured 2026-09-17T23:59:51Z was copied into the production directory as article.txt and IS the source of record for every narrated fact. Curly apostrophes and curly quotation marks in the snapshot were normalized to plain ones in the quotations below.)
• COVERS: SCENE 1 (two sentences). (a) EXACT NARRATED TEXT: "The article describes Cooley as an international law firm handling capital markets work. The firm says it developed a system called GO Public." (b) VERBATIM SUPPORT FROM article.txt: - line 51 (first two sentences): "Cooley is an international law firm with a reputation for helping companies navigate capital markets and initial public offerings (IPOs). In 2025, the firm advised on 180 deals globally, totaling more than $51.5 billion in deal volume." - line 53: "Cooley developed GO Public, a proprietary AI product offering built on ChatGPT Work." - line 52 attributes the practice context to a named executive: "explains David Wang, Chief Innovation Officer at Cooley." ONSCREEN CARD FOR THIS SCENE: "Cooley: 180 Deals $51.5B 2025". The card carries the two deal statistics with their attribution to Cooley intact and is read by eye only, never spoken, which satisfies rule 6 (prefer zero spoken numbers) while keeping the figures on the record. The card deliberately labels them as Cooley's deals rather than as anything the AI produced. (c) WORDING CHANGES AND WHY: - "with a reputation for helping companies navigate capital markets" became "handling capital markets work". Two reasons: "navigate" used this way is a figure of speech (rule 8), and "a reputation for" is promotional framing from a vendor page that the narration should not repeat as fact. - The sentence opens "The article describes" so that even the firm profile is attributed rather than asserted in our own voice (rule 16). - "initial public offerings (IPOs)" was dropped from this sentence because the hook already spells the term out and the parenthetical initialism is banned in narration (rule 9). - The third sentence of line 51 was NOT narrated at all. It contains three hyphenated compounds that rule 7 bans outright and it is pure reputational marketing that measures nothing. - "Cooley developed GO Public, a proprietary AI product offering" became "The firm says it developed a system called GO Public". "says" supplies the required attribution (rule 16), "proprietary AI product offering" was dropped for word budget and because "offering" is ambiguous against "initial public offering" in this script, and the original comma is gone (rule 3). - "developed" is retained verbatim from line 53 precisely because it is past tense and carries no launch, availability or pricing claim. - Scene 1 is twenty three words, under the thirty word renderer cap.(First party openai.com CUSTOMER STORY page carried in OpenAI's Customer Stories section (the section is listed in the site footer at article.txt line 120), titled "How Cooley is accelerating IPO work with ChatGPT" (line 3 and line 31) and datelined September 17 2026 (line 30). The page carries two "Contact sales" calls to action (line 33 and line 73) and the customer story metadata template reading "Company size: Enterprise" / "Region: North America" / "Industry: Services" / "Products: ChatGPT" (lines 34 to 41), which is how OpenAI marks vendor marketing collateral about one of its own customers. It is NOT a launch announcement, NOT a general availability announcement, NOT a benchmark result, and it contains NO independently verified claim and NO measured outcome: no time saved, no cost figure, no error rate, no count of offerings run through the product, and no before and after comparison. Every statistic on the page was supplied by the subject firm or by OpenAI. The live URL returns HTTP 403 to our fetcher, which is the normal response from openai.com, so the alert time snapshot stored at scripts/data/article-cache/964ae1dd0facf551.txt and captured 2026-09-17T23:59:51Z was copied into the production directory as article.txt and IS the source of record for every narrated fact. Curly apostrophes and curly quotation marks in the snapshot were normalized to plain ones in the quotations below.)
• COVERS: SCENE 2 (one sentence). (a) EXACT NARRATED TEXT: "Cooley says the system is built on ChatGPT Work and analyzes information so lawyers can review and validate it." (b) VERBATIM SUPPORT FROM article.txt: - line 53: "Cooley developed GO Public, a proprietary AI product offering built on ChatGPT Work. Its agentic harness analyzes information so lawyers can review and validate the work. The workflow comes together into a tailored starting point for an IPO." - line 60: "The harness provides a controlled workflow for agents that lays out which steps agents can perform automatically, where lawyers must review or validate the work, and how it all comes together." - line 41 confirms the product metadata: "Products: ChatGPT". (c) WORDING CHANGES AND WHY: - "Its agentic harness" became "the system". "Agentic harness" is unexplained jargon that a sixty second Short cannot define, and spending words defining it would breach the word budget (rule 1). The mechanism claim itself is unchanged. - "Cooley says" was prepended. Line 53 is written in the article's declarative voice, but it is a vendor page describing its own customer's product, so rule 16 requires that the narration report it as the firm's claim and not as verified behaviour. - "the work" became "it" purely for word budget; the referent is identical. - "review and validate" is a two verb pair taken verbatim from line 53. It is not a multi item list and carries no comma, so rule 4 and rule 3 are both satisfied. - The third sentence of line 53 ("a tailored starting point for an IPO") was not narrated: it repeats the same idea and contains the banned initialism (rule 9). - The phrase at line 60 that Wang calls his curated prior analyses was deliberately not carried. It is a hyphenated idiom, banned twice over by rule 7 and rule 8. - Zero commas, zero digits, zero spoken numbers. Nineteen words, under the thirty word cap.(First party openai.com CUSTOMER STORY page carried in OpenAI's Customer Stories section (the section is listed in the site footer at article.txt line 120), titled "How Cooley is accelerating IPO work with ChatGPT" (line 3 and line 31) and datelined September 17 2026 (line 30). The page carries two "Contact sales" calls to action (line 33 and line 73) and the customer story metadata template reading "Company size: Enterprise" / "Region: North America" / "Industry: Services" / "Products: ChatGPT" (lines 34 to 41), which is how OpenAI marks vendor marketing collateral about one of its own customers. It is NOT a launch announcement, NOT a general availability announcement, NOT a benchmark result, and it contains NO independently verified claim and NO measured outcome: no time saved, no cost figure, no error rate, no count of offerings run through the product, and no before and after comparison. Every statistic on the page was supplied by the subject firm or by OpenAI. The live URL returns HTTP 403 to our fetcher, which is the normal response from openai.com, so the alert time snapshot stored at scripts/data/article-cache/964ae1dd0facf551.txt and captured 2026-09-17T23:59:51Z was copied into the production directory as article.txt and IS the source of record for every narrated fact. Curly apostrophes and curly quotation marks in the snapshot were normalized to plain ones in the quotations below.)
• COVERS: SCENE 3 (one sentence). (a) EXACT NARRATED TEXT: "A Cooley partner says the system helps the firm move through intensive preparation faster before lawyers apply their judgment." (b) VERBATIM SUPPORT FROM article.txt: - line 57, attributed on that same line to a partner and co chair of Cooley's global capital markets group: "GO Public helps us move through the intensive preparation faster so lawyers and management teams can spend more time on the decisions that shape the transaction". The quotation is truncated at exactly that point because the article continues after a dash character that rule 7 forbids this file from reproducing. - line 63, same speaker: "It's to get to a strong starting point sooner, so our lawyers can spend more time applying judgment, challenging the disclosure and thinking strategically about the issues that matter most to the company." - line 65, pull quote from the same speaker: "ChatGPT Work allows us to move more quickly through intensive preparation and give valuable time back to management." (c) WORDING CHANGES AND WHY: - The speaker is narrated as "A Cooley partner" rather than by name. Rule 11: his surname is a pronunciation risk for the voice engine and a garble costs a re roll. His full name is carried on this scene's onscreen card ("Partner Dave Peinsipp Quoted"), which is read by eye and never spoken, so the attribution is preserved for the viewer. - "helps us move through the intensive preparation faster" is kept close to the source but reworded into third person and attributed out loud. The substance is unchanged: a claim of speed made by the firm, not a measured result. - Everything after the dash at line 57 was dropped. It is a three item list ("judgment, market experience and strategic thinking") which rule 4 bans in narration for speech rate reasons, and it sits behind a banned dash character. - "so lawyers and management teams can spend more time on the decisions that shape the transaction" was compressed to "before lawyers apply their judgment" for word budget, drawing the verb from line 63's "applying judgment". - Four idioms in this part of the article were deliberately NOT carried, per rule 8: the article's phrase for a preliminary pass at line 62, its phrase for the most important parts of the work at line 62, its slogan for better output sooner at line 63, and its phrase for an industry resistant to change at line 68. Each would have been a figure of speech and two of them are also hyphenated. - No spoken number, no digit, no comma. Nineteen words, under the thirty word cap.(First party openai.com CUSTOMER STORY page carried in OpenAI's Customer Stories section (the section is listed in the site footer at article.txt line 120), titled "How Cooley is accelerating IPO work with ChatGPT" (line 3 and line 31) and datelined September 17 2026 (line 30). The page carries two "Contact sales" calls to action (line 33 and line 73) and the customer story metadata template reading "Company size: Enterprise" / "Region: North America" / "Industry: Services" / "Products: ChatGPT" (lines 34 to 41), which is how OpenAI marks vendor marketing collateral about one of its own customers. It is NOT a launch announcement, NOT a general availability announcement, NOT a benchmark result, and it contains NO independently verified claim and NO measured outcome: no time saved, no cost figure, no error rate, no count of offerings run through the product, and no before and after comparison. Every statistic on the page was supplied by the subject firm or by OpenAI. The live URL returns HTTP 403 to our fetcher, which is the normal response from openai.com, so the alert time snapshot stored at scripts/data/article-cache/964ae1dd0facf551.txt and captured 2026-09-17T23:59:51Z was copied into the production directory as article.txt and IS the source of record for every narrated fact. Curly apostrophes and curly quotation marks in the snapshot were normalized to plain ones in the quotations below.)
• COVERS: SCENE 4, THE HONESTY BEAT (two sentences). (a) EXACT NARRATED TEXT: "This is OpenAI's own customer story about its own customer. It reports no time saved and no outside verification." (b) VERBATIM SUPPORT FROM article.txt: - line 29 and line 30: the byline and dateline are "OpenAI" and "September 17, 2026". The publisher is the vendor, not a third party. - line 33 and line 73: "Contact sales" appears twice on the page, once directly under the headline and once at the foot of the story. - lines 34 to 41: "Company size: Enterprise" / "Region: North America" / "Industry: Services" / "Products: ChatGPT". This is OpenAI's customer story metadata template. - line 120: the site footer lists "Customer Stories" as the Business section that this page belongs to. - line 60: "Cooley partners closely with OpenAI to combine subject matter expertise and AI engineering knowledge to ensure the right processes produce the right outcomes." (Hyphen in the original compound removed to comply with rule 7.) The two parties are commercial partners, which is why no part of this page is independent. - Absence support for sentence two: the body runs lines 44 to 71 and contains no time saved, no cost figure, no error rate, no count of offerings processed, no study, no benchmark and no third party evaluator. Line 69 says the product "is designed to help lawyers consider more information", which is a statement of intent. (c) WORDING CHANGES AND WHY: - "customer story" is the article's own category label from line 120, so scene 4 names the genre in the publisher's own terms rather than editorialising. - "no outside verification" is the literal statement of what is missing. Rule 8 forbids any figure of speech, so nothing rhetorical was used. - Nothing in this scene claims the product does not work. It states only what the document is and what it does not report. - Zero commas, zero digits, zero spoken numbers. Nineteen words, under the thirty word cap. SENTENCE BY SENTENCE AUDIT: 1. Hook sentence one, "Cooley says it built an AI system for initial public offerings." Sourced to article.txt lines 51, 53 and 59. 2. Hook sentence two, "OpenAI's writeup reports no measured results." Sourced to the absence of any outcome figure across lines 44 to 71, with the only figures on the page sitting at line 51 and describing the law practice. 3. Scene 1 sentence one, "The article describes Cooley as an international law firm handling capital markets work." Sourced to line 51 first sentence. 4. Scene 1 sentence two, "The firm says it developed a system called GO Public." Sourced to line 53 first sentence. 5. Scene 2 sentence one, "Cooley says the system is built on ChatGPT Work and analyzes information so lawyers can review and validate it." Sourced to line 53 first and second sentences, reinforced by line 60 and line 41. 6. Scene 3 sentence one, "A Cooley partner says the system helps the firm move through intensive preparation faster before lawyers apply their judgment." Sourced to line 57, reinforced by lines 63 and 65. 7. Scene 4 sentence one, "This is OpenAI's own customer story about its own customer." Sourced to lines 29, 30, 33, 34 to 41, 73 and 120. 8. Scene 4 sentence two, "It reports no time saved and no outside verification." Sourced to the absence described above, plus the intent wording at line 69 and the commercial partnership at line 60. 9 and 10. Endcard, two sentences of fixed channel boilerplate. It makes no claim about this story and needs no article line. SENTENCE BUDGET: - Word counts per narrated field: hook 17, scene 1 narration 23, scene 2 narration 19, scene 3 narration 19, scene 4 narration 19, endcard narration 17. - Arithmetic: 17 + 23 = 40. 40 + 19 = 59. 59 + 19 = 78. 78 + 19 = 97. 97 + 17 = 114. TOTAL NARRATED WORDS = 114, inside the target band of 112 to 118 and well under the hard ceiling of 132. - Sentence counts per field: hook 2, scene 1 has 2, scene 2 has 1, scene 3 has 1, scene 4 has 2, endcard 2. Arithmetic: 2 + 2 + 1 + 1 + 2 + 2 = 10 spoken sentences, under the ceiling of 12. The four scenes use 6 of their allowed 8. - COMMA COUNT ACROSS ALL NARRATED FIELDS: 0. - SPOKEN NUMBER COUNT: 0. No digit character appears in any narrated field either. - LONGEST SCENE NARRATION: scene 1 at 23 words, under the 30 word renderer cap. The other three are 19 words each. - durationS SUM: 11 + 11 + 11 + 12 = 45, inside the QA gate band of 30 to 50. These integers are cosmetic planning values the renderer never reads. - Dash audit: no em dash, no en dash, no double hyphen and no hyphenated compound appears anywhere in this file. The only hyphen characters present are inside the ISO date in storyDate and the ISO timestamp and the source URL slug, all of which are required literal strings. HONESTY CAVEAT: This is an OpenAI customer story published by OpenAI about an OpenAI customer. It is vendor marketing collateral: it sits in the Customer Stories section, carries the Enterprise / North America / Services / ChatGPT metadata template and carries two Contact sales calls to action. Every statistic in it was supplied by the subject firm or by the vendor and nobody independent measured anything. The watcher scored it 84 HOT at detection and 84 HOT unchanged on rescore, not deduped, with no regional demotion. The score reasons are newly detected 12.0h at detection (+40), source OpenAI (+30), terms chatgpt (+6), and the ChatGPT autocomplete wave at 14 hits (+8). THERE IS NO LAUNCH VERB CREDIT ANYWHERE IN THAT SCORE: the HOT rating is feed recency plus publisher weight plus an autocomplete wave signal and nothing else. The article announces nothing. It states no launch date, no availability, no pricing and no general availability claim, and the script therefore contains no launch verb in any field. Most importantly, the article reports NO measured outcome of any kind: no time saved, no cost reduction, no error rate, no number of offerings run through GO Public and no before and after figure. The only hard numbers on the page, 180 deals and more than 51.5 billion dollars of deal volume in 2025, describe Cooley's LAW PRACTICE and not the AI product, and the page never ties them to GO Public. They are carried on the scene 1 onscreen card labelled as Cooley's own deals, are never spoken, and no narrated sentence places them near a claim about the software. CORROBORATION STATUS: 0 transcripts. transcripts.json in this production directory holds an empty videos array because the host's transcript downloader build dates from April 2024 and every pull failed. That is our own tooling failure, it carries zero information about the story, and it is cited nowhere in this script. Nothing in the narration describes the story as unconfirmed on account of it. 0 timeline probes: timeline-probes.json holds an empty probes array and is cited nowhere. No third party outlet, publication or channel is named or linked anywhere in this file. All six source entries point at the single first party page, and every narrated fact comes from article.txt. DELIBERATE OMISSIONS: 1. The third sentence of line 51, on Cooley's record in the issuer side market over the past two decades. Omitted as reputational marketing that measures nothing, and it contains three hyphenated compounds banned by rule 7. 2. The 180 deals and 51.5 billion dollar figures as SPOKEN content. Omitted from narration under rules 5 and 6 and moved to an onscreen card with attribution intact, because two numbers spoken close together can put our transcriber into a repetition loop. 3. Line 52's colour quotation about lawyers always being busy and thousands of things needing to be done during an offering. Omitted as atmosphere, and the second half is hyperbole. 4. Line 54's pull quote that ChatGPT Work is changing legal work more effectively and more democratically. Omitted because it is a three part list that rule 4 bans and because it is a STRONGER claim than anything narrated. 5. Line 58 on how teams previously began from a precedent from a comparable company. Omitted for word budget. It is process background and supports rather than contradicts the narration. 6. Line 59's quotation about being the leading capital markets firm. Omitted as self promotion. 7. The phrase at line 60 for curated prior analyses, the phrase at line 62 for a preliminary pass, the phrase at line 62 for the most important parts of the work, the slogan at line 63, the phrase at line 68 for an industry resistant to change, and the phrase at line 70 about a wider industry change. All omitted under rule 8 as idioms, several also under rule 7 as hyphenated. 8. Lines 64 to 66 on executive attention during preparation. Omitted for word budget; scene 3 already carries the same speed claim from the same speaker. 9. Lines 68 and 69 on professional duties and defensible outcomes. Omitted for word budget; it qualifies nothing that is narrated. 10. Lines 70 and 71, the partner's vision statements about the future of capital markets practice and enormous potential beyond offerings. Omitted precisely because they are far STRONGER and more speculative than anything the script says. 11. Page furniture: navigation at lines 5 to 28, the loading and share controls at lines 42 and 43, the repeated section headings at lines 44 to 50, the calls to action at lines 72 and 73, the Keep reading block at lines 74 to 83 which points at three unrelated OpenAI items including a separate Astra for Law post, and the entire site footer at lines 84 to 162. All omitted as chrome. CONFIRMED EXPLICITLY: nothing omitted contradicts, qualifies, narrows or reverses anything narrated. Every omitted item is either unrelated page furniture, colour, a list that would overrun the render, or a claim STRONGER than what the script narrates. There is no negative result, no caveat, no limitation, no correction and no dissenting voice anywhere in the snapshot that the script has hidden. PROMPT INJECTION CHECK: All 162 lines of the snapshot were read in full and treated as untrusted data that informs facts only and never instructions. Page furniture found: a top navigation and login bar (lines 5 to 28), the byline, dateline, headline, subtitle and a Contact sales button (lines 29 to 33), the customer story metadata template (lines 34 to 41), loading and share widgets (lines 42 and 43), section headings repeated three times by the page's sticky table of contents (lines 44 to 50), a second Contact sales button and a Try the new Astra for Law promotion (lines 72 and 73), a Keep reading block of three unrelated OpenAI items with their own datelines (lines 74 to 83), and a full site footer of product, research, developer, company, legal and locale links (lines 84 to 162). Imperative verbs do appear in that furniture: Skip to main content, Log in, Try ChatGPT, Contact sales, Share, Try the new Astra for Law, Keep reading, and Your privacy choices. Every one of them is ordinary website interface copy addressed to a human visitor. RESULT: NO injection found. No imperative anywhere in the file was addressed to an AI system or assistant. Nothing attempted to alter a system prompt, a persona, a role or a set of rules. Nothing requested a tool call, a file write, a network fetch, a credential or a command. No hidden payload, no encoded block, no invisible text marker, no fake system message and no instruction disguised as a quotation was present. Nothing from the snapshot was treated as an instruction at any point.(First party openai.com CUSTOMER STORY page carried in OpenAI's Customer Stories section (the section is listed in the site footer at article.txt line 120), titled "How Cooley is accelerating IPO work with ChatGPT" (line 3 and line 31) and datelined September 17 2026 (line 30). The page carries two "Contact sales" calls to action (line 33 and line 73) and the customer story metadata template reading "Company size: Enterprise" / "Region: North America" / "Industry: Services" / "Products: ChatGPT" (lines 34 to 41), which is how OpenAI marks vendor marketing collateral about one of its own customers. It is NOT a launch announcement, NOT a general availability announcement, NOT a benchmark result, and it contains NO independently verified claim and NO measured outcome: no time saved, no cost figure, no error rate, no count of offerings run through the product, and no before and after comparison. Every statistic on the page was supplied by the subject firm or by OpenAI. The live URL returns HTTP 403 to our fetcher, which is the normal response from openai.com, so the alert time snapshot stored at scripts/data/article-cache/964ae1dd0facf551.txt and captured 2026-09-17T23:59:51Z was copied into the production directory as article.txt and IS the source of record for every narrated fact. Curly apostrophes and curly quotation marks in the snapshot were normalized to plain ones in the quotations below.)