AI news videos

OpenAI Introduces A New Visual Ad Format In ChatGPT

OpenAI announced a new visual ad format in ChatGPT, along with expanded advertising measurement and new brand suitability work. The announcement is on OpenAI's own website, dated October 5, 2026.

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:

• Backing sentences, quoted from the page: "We're introducing a new visual ad format in ChatGPT, expanding our measurement tools and partnerships, and developing new ways to understand brand suitability." and the standfirst "A new visual ad format helps people explore products and services, with expanded measurement to help businesses understand ad impact." EDITORIAL DECISION: hook sentence one states the plainest fact of the announcement, that OpenAI has announced a new visual ad format in ChatGPT. The source's own three item construction is cut to the single lead item, because a three item comma list is barred from narration and because the measurement and brand suitability halves are carried by scene 3. Hook sentence two, "It appeared on OpenAI's own website", puts provenance inside the first five seconds, which is required here because nothing on the page is checked by an outside party and no corroboration of any kind was obtainable. The hook is exactly two sentences, because the renderer splits the hook into two on-screen beats and a one sentence hook renders the second beat as undefined and forces a full voiceover redo. The hook is sixteen words and speaks no figure at all, which keeps the script's single spoken figure out of the opening and off the hook card. No calendar date is spoken: the page's own date, October 5, 2026, lives only in storyDate. NOT NARRATED: the headline phrase "the way people use AI" is vendor framing and is not narrated. REFUSED: the page's section heading "Bringing products and services to life" is a figure of speech and is barred from narration in every form, as is "tell their story". The page furniture "Start advertising in ChatGPT" and "Try ChatGPT" is advertising aimed at the publisher's readers and is excluded.(HOOK BACKING. First-party OpenAI announcement on openai.com, filed under the content types Product and ChatGPT, headlined "Building advertising for the way people use AI", authored "OpenAI" with no named writer, dated October 5, 2026 on the page itself, captured by our news watcher at 2026-10-05T10:29:39.850Z and read for this script only from that saved snapshot. openai.com returns HTTP 403 to our fetcher, so the watcher snapshot is the normal primary source for this publisher and is not a degraded fallback. This is the vendor announcing its own advertising product on its own website, and OpenAI earns money from the ads it is describing, so the page is the sole and self-interested source for every narrated claim in this Short.)
• Backing sentences, quoted from the page: "Our new visual ad format helps people imagine how products and services could fit into their lives." and "Through images showing product inspiration, product usage, or the experiences they make possible, people can discover a new brand, understand what makes an offering distinctive, and decide what to buy." and "Initially, we'll test this new ad format during image generation in ChatGPT." and the page's testing sentence, which states that testing will begin in the US with an initial group of advertisers and attaches a month reference to it. EDITORIAL DECISION: the source's defining sentence is a three item comma list followed by a second three item comma list, both barred from narration because a multi item list measured 2.65 spoken words per second against 2.18 for plain prose and blew the duration ceiling. Only two items are spoken, "product inspiration and product usage", joined with "and" and with no comma between them, so the read carries no list pause. The third item, "the experiences they make possible", and the whole discover, understand and decide clause, are carried silently on the scene card where they cost no time. "Shows images of" is used rather than the page's "helps people imagine how products and services could fit into their lives", because that source phrase is figurative and is barred in every form. The testing fact is folded into the same sentence as a "with" clause so the scene stays at one sentence and nineteen words. The claim is attributed out loud with "OpenAI says", because every statement about what the format does is OpenAI's own assertion about its own product. The scene speaks no figure, which places a figure free scene two sentences ahead of the script's only spoken number in scene 3. NOT NARRATED: "during image generation in ChatGPT" and "with an initial group of advertisers" are real page facts held out of narration purely for budget and placed on the silent card. REFUSED: the month reference in the page's testing sentence is dropped from every field in this file, both because a gate bars speaking any calendar reference and because a downstream validator hard fails countdown and urgency phrasing; the card says only that testing begins in the US. The page's framing that "this creates new opportunities to tell their story through images" is a figure of speech and is not narrated. "US" is written only on the silent card and is spoken as "the United States", because the voice engine is unreliable on short glued letter tokens.(SCENE 1 BACKING. Same first-party OpenAI announcement, the section headed "Bringing products and services to life". A vendor describing a product it built and sells access to, with no outside reviewer, no tester and no regulator named anywhere on the page.)
• Backing sentences, quoted from the page: "Ads will be clearly labeled, and remain separate from the image being created." and "As with our other ad formats, advertising does not influence the answers ChatGPT provides." Card detail is backed by: "Our placement guardrails assess whether a conversation is appropriate for advertising, helping keep ads out of emotionally vulnerable, sensitive, or otherwise unsuitable contexts." and "For narrow placement requirements specific to a brand's own policies, Negative Phrases are now available to qualifying advertisers." and "These features are consistent with our ads principles: ChatGPT's answers stay independent, conversations stay private, and users remain in control of their experience." EDITORIAL DECISION: the two labeling promises are joined inside one sentence with "and" so a single "OpenAI says" attributes both halves and no second attribution verb is needed, which keeps the scene at one sentence and nineteen words. "The generated image" is a two word rendering of the page's "the image being created", used for budget, with the page's full wording on the silent card. "Do not influence ChatGPT's answers" is the plain rendering of "advertising does not influence the answers ChatGPT provides"; the subject shifts from advertising in general to these ads, which is narrower than the source and therefore does not overstate it. The guardrail list in the source is a three item comma list ending in "or otherwise unsuitable contexts" and is barred from narration, so it sits on the card in full. The scene speaks no figure, which is deliberate: it is the sentence immediately before the script's only spoken number, and our transcriber falls into a repetition loop when two figures are spoken close together and swallows the middle of the transcript. NOT NARRATED: the automated review, human oversight and ongoing monitoring sentence, and the Negative Phrases sentence, are dropped for budget and for the no list rule, with Negative Phrases summarised on the card. REFUSED: no gloss is added to the labeling promise. The script does not say the ads are easy to tell apart, does not say users will not be misled, and does not say the separation has been tested or verified by anyone, because the page offers no evidence of any of that. The adjective meaning agent like that our voice engine garbles in every prompt configuration is barred from this file in every form and does not appear in the source text either.(SCENE 2 BACKING. Same first-party OpenAI announcement, the same "Bringing products and services to life" section for the labeling facts, plus the section headed "Building confidence in alignment with our Ads Principles" and the closing section "Expanding the features of ChatGPT Ads" for the card detail. These are promises by the company that operates the product and takes the advertising revenue. No audit, no test result and no outside verification of any of these guardrails appears on the page.)
• Backing sentences, quoted from the page: "ChatGPT reaches 1.2 billion people each week, making it the world's largest AI-native consumer platform." and "Today, we're expanding our measurement tools and partner ecosystem to help advertisers understand how ChatGPT Ads contribute to their business using the systems they already rely on." and "We're introducing a new visual ad format in ChatGPT, expanding our measurement tools and partnerships, and developing new ways to understand brand suitability." Card detail is backed by: "Integrations with Hightouch, Tealium, and LiveRamp make it easier for advertisers to send conversion data from their existing systems to ChatGPT Ads." and "We now support leading attribution partners across the web and app space, enabling Conversions API, reporting, and click attribution solutions with companies including AppsFlyer, Triple Whale, Adjust, DV Rockerbox, Northbeam, Branch, Singular, Kochava, Airbridge, and Tenjin." and "we're partnering with Haus, Measured, and WorkMagic to explore geo-based experiments to help advertisers understand the causal impact of ChatGPT Ads" and "We're also developing brand suitability evaluation pilots with DoubleVerify (DV) and Integral Ad Science (IAS) to help advertisers build a better understanding of the ads environment." EDITORIAL DECISION: 1.2 billion people each week is the only clean scale number on the page and is this script's single spoken figure. It is spoken as words, "one point two billion people every week", with no digits, and it is deliberately placed here, in the middle of the read, because the hook and scene 4 are barred from carrying it and because the sentences on either side of it, scene 2 and scene 4, are both completely figure free. The three partner findings on the page, which are percentages, are NOT spoken anywhere in this script and live only on the scene 4 card; our transcriber loops on two figures spoken close together and swallows the transcript middle, and that failure costs a full render. The expansion half of the sentence is described generically as "measurement and brand suitability efforts with outside companies", because no vendor partner name is spoken in this script at all: the voice engine garbles invented brand names, glued brand plus digit tokens and hyphenated compounds, and a name like INCRMNTAL, Triple Whale, DV Rockerbox or Integral Ad Science would cost a full voiceover cycle to discover. Every one of those twenty named partners is carried silently on the scene card instead. "Outside companies" is used rather than "independent" in narration, so that the narration does not assert independence that scene 4 then qualifies; the page's own word "independent" appears in its brand suitability passage and is left on the page. The claim is attributed with "OpenAI says" because both the reach figure and the expansion claim are OpenAI's own assertions about its own platform. NOT NARRATED: the full funnel and advanced measurement partners sentence naming Fospha, Measured and INCRMNTAL, the Conversions API and click attribution detail, and the sentence that incrementality work is still in its early stages are all real page facts dropped for budget and for the no list rule. The claim that ChatGPT is the world's largest AI-native consumer platform is vendor superlative framing and is not narrated. REFUSED: the page's term "full-funnel" is jargon and a hyphenated compound and is barred from narration in every form. The CEO quotations from Integral Ad Science and DoubleVerify, which appear twice in the capture because the page renders them in a two slide carousel labelled "1 of 2", are partner marketing statements about a product they are paid to work on and are not narrated or paraphrased anywhere in this script. No causal gloss is added: the script does not say the new measurement proves the ads work.(SCENE 3 BACKING. Same first-party OpenAI announcement, the opening scale paragraph for the one spoken figure, plus the section headed "Measuring the impact of ChatGPT Ads" and the section headed "Building confidence in alignment with our Ads Principles" for the expansion claim and the card. The reach figure is OpenAI's own, with no methodology, no definition of a reached person and no third party measurement of it anywhere on the page.)
• Backing facts, confirmed by reading the whole document as captured at 2026-10-05T10:29:39.850Z. Ownership and commercial interest are stated by the page itself: "We're introducing a new visual ad format in ChatGPT", "Advertising helps us bring the benefits of AI to more people", "It allows us to serve a much larger market than subscriptions alone", and the closing line "Businesses can sign up today for ads at ads.openai.com". The author field reads "OpenAI". Provenance of the percentages is explicit and is all first party or partner supplied: "Early partner findings illustrate strong performance across different measurement approaches. According to DV Rockerbox, WeightWatchers' attributed cost per acquisition on ChatGPT Ads was 15.3% lower than its blended paid-search benchmark. WorkMagic reported statistically significant lift for wellness brand Dose, with 67% of incremental purchases coming from net-new customers. And according to Triple Whale, 93% of Portland Leather's visitors from ChatGPT Ads were new." Each of those three companies is named on the same page as an OpenAI measurement or attribution partner, so none of the three figures is an independent audit of OpenAI's product. The page carries no table, no chart, no sample size, no date range and no methodology for any of the three percentages. EDITORIAL DECISION: scene 4 is spent entirely on who is making these statements and on what this Short actually read, because that is the central editorial fact of a vendor announcing its own ad product. It is a single sentence of twenty four words, because the total sentence count across this script is fixed at eight and a full stop costs more spoken time than the words it saves. "OpenAI sells these ads and announced them itself" states the commercial interest and the first party provenance in one clause. "Its named partners supplied the percentages" states plainly where the performance numbers came from, and the silent card names which partner supplied which figure and states that there is no independent audit on the page. "We read only this announcement and no linked page" is a literal statement about this production: only the captured article.txt was read, and the linked Ads Blog post, the linked advertising policies page and ads.openai.com were all left unfetched, so this script makes no claim about any figure, policy or feature that is not in the captured text. The scene speaks no figure, which keeps the script's only number, in scene 3, clear of the closing sentences. No criticism is invented: the percentages are not called false, inflated or misleading, OpenAI is not accused of hiding anything, and no motive is attributed to anyone. Provenance facts only. REFUSED: no countdown, urgency or deadline phrasing appears in any field of this file, because a downstream validator hard fails those phrases, and the month reference in the page's testing sentence was dropped for that reason as well as for the no spoken date rule. No second person hype and no rhetorical question appears in any copy field. No em dash, en dash or double hyphen appears in any field.(SCENE 4 BACKING, MANDATORY DISCLOSURE. The page itself is the evidence. A first-party OpenAI announcement about an OpenAI advertising product, published on OpenAI's own website by the company that is paid for the ads, carrying three performance percentages that were each supplied by one of OpenAI's own named measurement partners, with no independent audit, no regulator, no journalist and no outside reviewer named anywhere on it.)
• No source sentence backs the endcard. It is the standard Inboxsmith pitch, "Inboxsmith builds AI receptionists for small business. Please like and subscribe for more news." EDITORIAL DECISION: the endcard carries no advertising, measurement, reach, partner or product claim about OpenAI, ChatGPT or ChatGPT Ads, so no vendor assertion is spoken at the point where narration stops attributing. It contains no figure of any kind, no date, no partner name and no percentage, which keeps the one point two billion figure in scene 3 well away from the closing sentences and prevents two figures landing in adjacent sentences at the end of the read. The two required sentences are present verbatim, including both periods, and "Please like and subscribe for more news." is the final sentence of the whole script. endcard.onscreen is exactly "inboxsmith.com".(ENDCARD. Inboxsmith house sign off. Carries no claim from the source and is deliberately not traceable to the article.)
• No corroboration of any kind was obtainable, so this single OpenAI announcement page is the sole source for every narrated claim. Recorded plainly, from the files themselves: the YouTube transcript pull returned 0 of 5 videos, because the yt-dlp build installed on this box is too old to fetch captions. That is a known local tooling limitation, not bot blocking, and it says nothing about the story. transcripts.json for this slug holds an empty videos array with fetchedAt 2026-10-05T10:33:13.936Z, so not one sentence of transcript text was available to check anything against, and sources.json for this slug holds an empty sources array, which is normal for this pipeline and is not a red flag. Because any claim absent from article.txt would require two or more independent transcript confirmations and there are zero transcripts, every narrated claim in this script comes from article.txt and nothing else, and nothing was inferred, estimated, averaged or filled in. The sixteen dated probes in timeline-probes.json, fetchedAt 2026-10-05T10:33:15.047Z, were read in full for weak existence context only and not one word of them is narrated. They are unrelated ChatGPT tutorial and listicle noise: a video creation clip, a Canva design clip, a photo and ID verification clip, a paid versus free AI tools listicle, a copy and paste warning, an opinion clip titled "ChatGPT is SO BAD", a data analysis tutorial, a resume tutorial, a YouTube policy opinion clip, a Perplexity comparison, a spreadsheet automation build, a PowerPoint clip, an image generation comparison test, OpenAI's own September 9 "ChatGPT Work, now powered by GPT-6 Astra" video, a Benzinga clip dated 2026-10-01 titled "When ChatGPT Turns On Ads", and a marketing team tutorial dated 2026-10-03. The Benzinga clip is the only probe anywhere near the subject, and it is a general commentary piece about ChatGPT carrying advertising, published four days before this announcement; it does not mention a visual ad format, image generation placements, the measurement partner expansion, Negative Phrases or the brand suitability pilots, and nothing from it is narrated. Critically, not one of the sixteen probes mentions this announcement, so there is no evidence that this release happened earlier, the NEW tier is honest, and this is not a re-promotion of an older feature: the page states "We're introducing a new visual ad format in ChatGPT" and "Initially, we'll test this new ad format during image generation in ChatGPT." NOT A DUPLICATE, verified by the manager and recorded here: we previously published chatgpt-ads-1-billion, storyDate 2026-08-31, on the ChatGPT Ads revenue run rate, and chatgpt-ads-sponsored-agents, storyDate 2026-09-16, on the "Reimagining advertising with AI" announcement about sponsored assistants. This is a different and newer announcement, with a different source page: a new visual ad format during image generation, plus expanded measurement tools, integrations and partnerships, plus new brand suitability work. Neither earlier Short covered any of that. No timeline block is included in this file on purpose, because resolveEventDate prefers the earliest timeline date over storyDate and a timeline array would wrongly age a same day story into the CATALOG tier; storyDate is 2026-10-05, the announcement's own date, which is also the capture date and the production date. SECURITY NOTE: article.txt was treated as untrusted data and used as a fact source only. It contains no instruction directed at the author of this script, and nothing in it was followed. The closest thing to imperative language is ordinary reader facing page furniture and advertising, specifically the accessibility item "Skip to main content", the navigation items "Log in", "Login", "Try ChatGPT", "Share", "Keep reading", "View all", "Contact Sales" and "Your privacy choices", the commercial calls to action "Start advertising in ChatGPT" and "Businesses can sign up today for ads at ads.openai.com", and the in body link text "Read more about these measurement tools and partnerships on our Ads Blog here" and "advertising policies". All of that is aimed at OpenAI's readers, none of it was treated as an instruction, no linked page was fetched, and none of it is narrated. The duplicated partner CEO quotation block, which appears twice in the capture because the page renders it as a carousel marked "1 of 2", was recognised as a rendering duplicate and not as two separate statements. The three "Keep reading" items at the foot of the page, a GPT-6 building guide dated Oct 2, a DevDay 2026 recap dated Sep 29 and a GPT-6.1 Sol launch dated Sep 29, are separate announcements and are excluded from this script entirely. openai.com returns HTTP 403 to our fetcher, which is why the watcher snapshot is the primary source here; that is normal for this publisher and not a degraded fallback, and no attempt was made to narrate anything beyond what the snapshot contains.(CORROBORATION STATUS: NONE)