If your business shows up in ChatGPT mainly because news outlets have written about you, that visibility now rests on ground that moves without warning. A ChatGPT model launch this summer cut citations to major news sites, and separate tracking shows brands' own websites and official sources taking a larger share of what ChatGPT cites. The practical response is simple: put the facts you want repeated on your own site, and check how ChatGPT describes you every time a major model launches.
The July drop was a ChatGPT change, not a Google one
On July 9, 2026, OpenAI rolled out the GPT-5.6 family in ChatGPT. Seer Interactive independently confirms that GPT-5.6 Luna became the default model that day. Within a week, ChatGPT's citations of the Associated Press, TIME, TechCrunch, the Los Angeles Times and The Atlantic fell by 50% to 90%. That figure comes from Muck Rack's Generative Pulse research, presented by Generative Pulse's Matt Dzugan at its October 28 summit and reported by Anna Crowe in Search Engine Land on November 16, 2026. Crowe notes that Generative Pulse queries AI models through their APIs rather than the apps customers use, and tells readers to "Read it as direction, not gospel."
Crowe also points out what did not happen. In her account, nobody changed robots.txt, Google made no algorithm change and OpenAI posted no changelog. The drop lined up with a model launch day. OpenAI pays AP, TIME and The Atlantic for journalism, yet outlets with no public OpenAI deal, including Reuters, CNBC, TechCrunch and the Los Angeles Times, fell too.
She then checked whether those outlets had lost ground in Google at the same time. Her own Semrush analysis of keywords ranking in Google's top three, from June 15 to July 15, showed mixed moves: AP up 17%, TIME up 1%, TechCrunch up 3%, the LA Times down 2% and The Atlantic down 8%. Nothing there explains a 50% to 90% fall in ChatGPT citations. The reverse also held. When a May Google core update cut top-three keyword counts for TIME, TechCrunch, the LA Times and The Atlantic by 17% to 22%, their ChatGPT citations "held flat through June." Gemini's citations of the same outlets, she adds, "barely moved." She treats July as a ChatGPT change, not a Google one.
Two disclosures matter when you weigh this. Crowe works at Muck Rack, which produces Generative Pulse and hosted the summit where the data was presented, and Search Engine Land is owned by Semrush. She says the Semrush analysis is her own.
For a business owner, the lesson is that your AI visibility can change within a week of a model launch, even when your website and your search rankings have not changed.
ChatGPT now goes straight to the source
So where did those citations go? Crowe's phrase for it is that "ChatGPT's search now goes straight to the source, and news is the article about the source."
Several measurements point the same way. They measure different things, so it helps to keep them apart.
First, Muck Rack's October 28, 2026 Generative Pulse release says journalism fell from 28.1% of ChatGPT's citations in January to 22% in September, "with sharp drops in the spring and July." Crowe's article gives roughly 33% as the January starting point; the release is the primary source. Claude and Gemini, Crowe writes, "mostly sat still."
Second, Generative Pulse asked ChatGPT the same 102 questions in April and again in September. In that panel, as Crowe reports it, government and NGO sources rose from 10% to 35% of citations, brands' own websites rose from 14% to 20%, and journalism fell from 28% to 22%. Crowe flags that the September run used a newer ChatGPT model, "so it reflects the model update as much as the calendar."
Third, the newer model searches differently. In a September 24, 2026 study of 346 matched prompts, Seer Interactive found that GPT-5.6 Luna used site: searches, which look inside one specific website, in 76% of its fan-out queries, against 21.6% for GPT-5.5. It also consulted 43% fewer sources, 3,481 against 6,134. Seer sells AI visibility consulting, so treat this as a vendor study. Ahrefs, as cited by Crowe, found that when Luna ran site: searches in logged-out ChatGPT against the most-searched domains, 75.9% targeted brand sites.
Put plainly, ChatGPT is increasingly skipping the story about you and reading your own pages instead. That is a risk if your site is thin. It is an opening if your site says clearly what you do.
Be the thing the article is about
Crowe's practical advice is in her own words: "Be the thing the article is about." She recommends original data, real documentation and a newsroom a crawler can read. She also reports that when ChatGPT cites a press release, "it links to the company's own site about 85% of the time and to a wire only 14%." The brands she expects to be hurt most are those that depend on one type of source, such as news coverage, to appear in AI answers.
Your own site is not the whole answer. Muck Rack's release says earned and third-party sources still make up more than 84% of AI citations across the four AI providers it tracks, and that "brands can't control their AI visibility through their own websites alone." Keep earning coverage and reviews. Just don't let them be the only place your facts live.
Media deals do not solve this either. Crowe cites a June OtterlyAI study that found OpenAI licensees earned 48% more ChatGPT citations per page, but that was before July 9. Her verdict: "A deal may raise your floor. It doesn't stop the drop."
For a privately held business, our advice is to make sure four things live on your own site, in plain text a model can quote:
- Who you serve. The industries, regions and customer types you work with, stated directly.
- Your pricing. Real numbers or clear ranges where you can publish them, and how pricing works where you cannot.
- Your policies. Returns, service terms, response times, security and data handling.
- Your proof. Case studies, credentials and original data that someone else could cite.
These pages are also what an AI assistant needs when it acts for a customer, not just when it answers a question. We covered that side in Your next customer might be a bot.
Your own site is not the only surface ChatGPT reads. If you have not yet claimed your ChatGPT business profile, start with Businesses in ChatGPT: How to Claim Your Business Profile and What It Does. If you have, make sure the listing describes you correctly and matches what your site says.
Recheck at every model launch, not once a year
The July drop was not a one-off. Crowe lines up earlier waves with ChatGPT launches. GPT-5.4 Thinking launched in ChatGPT on March 5, and by March 9 the Financial Times had vanished from its citations. GPT-5.5 Instant became ChatGPT's default the week of May 5, six days before a wave that hit outlets including The Verge, The Wall Street Journal and Politico. Two other waves, in January and June, matched no launch she could find.
SISTRIX, which calls these shifts "ChatGPT Core Updates," measured one directly. In a June 1, 2026 post, it reported that after ChatGPT switched from GPT-5 mini to GPT-5.5 on May 23, the mix of domains cited in German-language responses shifted 47% within 48 hours, against a usual day-to-day drift of 1% to 2%. Its sample was 3.8 million German-language responses. SISTRIX is careful to call this a correlation: from the outside, it cannot prove the model alone caused the change.
That pattern changes how often you should look. A once-a-year check will miss a launch-week swing. A better habit is to check how ChatGPT answers questions about your business in the days after each major model launch, and give one person on your team clear ownership of that check so it does not fall between remote schedules.
Be careful with single answers, too. Muck Rack's release reports that when the same prompt ran twice in one day, ChatGPT repeated just 41% of its cited domains, against 45% for Gemini and 94% for Perplexity. Crowe adds that of more than 20,000 sites cited by the four models, only about 200 to 250 appeared in all four, and she calls a single "AI visibility score" that averages four models a vanity metric. Look at the actual answers to the questions your customers ask, model by model, and look more than once.
For a starting checklist, see Is Your Business Invisible to ChatGPT?. Google's AI answers deserve the same attention, and for many buyers they come first. Our companion piece, Before you chase ChatGPT, fix how Google's AI describes you, covers that side, and Google AI Overviews and Answer Engine Optimization covers the mechanics.
Sources
- Anna Crowe, "ChatGPT citation drops look like unannounced Google core updates," Search Engine Land, November 16, 2026: https://searchengineland.com/chatgpt-citation-drops-unannounced-google-core-updates-501912. Figures from Semrush, Ahrefs and OtterlyAI are as cited by Crowe.
- Muck Rack, "Generative Pulse: Earned Media Still Drives More Than 84% of AI Citations," GlobeNewswire, October 28, 2026: https://www.globenewswire.com/news-release/2026/10/28/3390134/0/en/generative-pulse-earned-media-still-drives-more-than-84-of-ai-citations.html. Same research program Crowe reports on.
- Seer Interactive, "AI Citations Aren't Your Top KPI: ChatGPT-5.6 Luna Runs More Targeted Searches," September 24, 2026: https://www.seerinteractive.com/insights/gpt-5.5-vs-gpt-5.6-luna
- SISTRIX, "ChatGPT Core Updates: Chatbots Have Algorithm Changes Too," June 1, 2026: https://www.sistrix.com/blog/chatgpt-core-update/
We can help you with that.
NTx AI Consulting works with owner-led and partner-led privately held businesses that are currently using AI tools in everyday workflows, or would like to be. The complimentary consultation is a working session, not a product demo. If a deeper engagement makes sense, Phase 1 is an AI Disruption Readiness Audit. You can learn more at https://ntx.ventures/ai-consulting/.