Half of everything AI Search cites is a page a company wrote about itself. Journalism is one citation in six. And the split between them swings by a factor of seven depending on which industry you are in, which means the GEO advice most teams are following is written for an average sector that does not exist.
We know because we asked all seven AI Search engines the same thousands of questions every day through July and August 2026, in each of 17 industries (US only), written the way buyers actually ask: ranked lists, comparisons, pricing, how to questions and trust checks. That produced 6,095,520 citations on clean collection days, plus 107,633 ChatGPT ad units and 15,363 shopping cards, each one classified into one of ten source categories and tagged to its industry and engine.
It also caught something we were not looking for. On 8 August, and again on 14 August, ChatGPT changed what it cites. No other engine made the same move. This report covers both.
Every figure is live and filterable at interactive The AI Search Index: Industry Report
Key Findings
- Company owned pages take 53.7% of all citations. News and media takes 16.9%.
- Journalism dependence swings 7.3x by sector, from 32.2% in Technology to 4.4% in AI Data Governance.
- News and Media is the second largest category by AI Citations in 17 Industries
- Educational Platforms draws 30.7% from education domains, against a median sector of 1.1%.
- On 8 August ChatGPT halved its journalism, from 21.3% to 9.8%, while its citation volume doubled.
- Community and forum sources fell 67% on ChatGPT across the same break, then fell again on 14 August to end 80% down.
- Five other engines held within 7%, so this is ChatGPT, not AI Search.
The Most Cited Sources on AI Search
Company owned pages take 53.7% of every citation the seven engines make. News and media takes 16.9%, government and NGO sources 8.1%, and nothing else clears 5%.
The debate about AI Search is about publishers, and publishers are real. But one citation in six is journalism, and more than half is a page some company wrote about itself.
The August Shift: ChatGPT 5.6’s Impact on The AI Search Landscape
Following the rollout of ChatGPT 5.6 (GPT-5.6) changes started happening. On 8 August, ChatGPT’s citation volume roughly doubled, from a median of 19,801 a day to 39,446, but what it cited also changed.
| Category | Baseline, Jun – Aug | After 14 August | Change |
|---|---|---|---|
| Company owned | 44.38% | 55.28% | +25% |
| News and media | 21.33% | 10.40% | -51% |
| Government and NGO | 13.87% | 17.78% | +28% |
| Community and forum | 4.79% | 0.97% | -80% |
| Blogs and personal sites | 3.20% | 2.24% | -30% |
| Other | 5.39% | 4.22% | -22% |
| Education | 4.95% | 4.84% | -2% |
| Encyclopedia | 1.45% | 1.43% | -1% |
| Social media | 0.26% | 0.84% | +223% |
| Video | 0.33% | 0.98% | +197% |
The doubling and the redistribution are the same event. ChatGPT started citing roughly twice as many sources per answer, and the extra sources it reached for were overwhelmingly company owned and official, not journalistic.
No other AI Search Engine did the same
This is what makes the finding safe to publish. If AI Search as a whole had shifted, every engine would show it. Across the same break:
| Engine | News before | News after | Change |
|---|---|---|---|
| Perplexity | 22.7% | 23.4% | +3% |
| Microsoft Copilot | 14.7% | 14.2% | -3% |
| Google AI Mode | 18.3% | 17.6% | -4% |
| Claude | 20.1% | 18.8% | -7% |
| Google AI Overviews | 14.0% | 13.0% | -7% |
| Gemini | 16.1% | 14.0% | -13% |
| ChatGPT | 21.3% | 10.4% | -51% |
Five engines moved between -7% and +3%. ChatGPT moved -51%. Six control groups running the same prompts on the same days is about as clean a natural experiment as this field offers.
Gemini is the one engine worth a second look, and it is not a second ChatGPT. Its news share fell to 4.8% between 18 and 22 August, then came all the way back: 16.4% on 24 August, and holding between 14% and 18% for the rest of the month. Its citation volume never dropped, so this is not a collection gap. Excluding those five days Gemini is down 4%, in line with the other controls. ChatGPT changed level and stayed there. Gemini wobbled and recovered.
How Did 9 Source Category Behave After The ChatGPT 5.6 Update?
1. Company Owned Content
Company Content in ChatGPT
44.38% before the break, 55.28% after, +25%. The winner. Company pages absorbed most of what journalism and forums lost, and at 53.7% across all seven engines this is already the largest surface in AI Search.
The range is enormous
Industrial/Manufacturing draws 77.4% of its citations from company websites. Nonprofit draws 29.6%. That is the difference between a category where your own site is nearly the whole game and one where it is a minority voice.
Company Content in all 7 AI Search Engines
2. News and Media
News and Media in ChatGPT
21.33% before, 10.40% after, -51%. News websites are the largest loser of the GPT 5.6 changes, and the cleanest step in the whole dataset. Across all engines and sectors journalism still holds 16.9% of citations, but its dependence runs from 32.2% in Technology to 4.4% in AI Data Governance, a spread of 7.3x.
Which newspapers ChatGPT stopped citing
The fall is not evenly spread. Taking the six most cited outlets and measuring each as a share of all ChatGPT citations, every one of them steps down on 8 August:
| Outlet | Share of the category | Before | After | Change |
|---|---|---|---|---|
| forbes.com | 7.9% | 1.775% | 0.708% | -60% |
| techradar.com | 3.8% | 0.921% | 0.214% | -77% |
| nerdwallet.com | 3.1% | 0.668% | 0.363% | -46% |
| tomsguide.com | 2.5% | 0.511% | 0.324% | -36% |
| reuters.com | 2.4% | 0.655% | 0.233% | -64% |
| wsj.com | 2.4% | 0.486% | 0.208% | -57% |
No outlet was spared and none was singled out. Tech review sites fell hardest and the wire services and business press fell nearly as far, which points at a change in how ChatGPT weights journalism as a class rather than a judgement about particular publishers.
News and Media in all 7 AI Search Engines
3. Government and NGO
Government and NGO in ChatGPT
13.87% before, 17.78% after, +28%. Regulators and NGOs gained almost as much as company pages did.
Two sectors run on it
Nonprofit draws 42.4% of its citations from government and NGO sources, ten times the median sector, and Pharma 26.2%. In both, the team that controls AI visibility may not report to the CMO. Regulatory affairs, clinical publishing and grant reporting are GEO functions there whether anyone has told them so or not.
Government and NGO in all 7 AI Search Engines
4. Community and Forum
Reddit loses over -80% in in ChatGPT
4.79% before, 0.97% after, -80%. The steepest fall of any category, and it happened in two stages: down 67% on 8 August, then again on 14 August.
reddit.com alone ran at 4.42% of all ChatGPT citations before August and 0.39% after 14 August, a fall of 91%. Every other forum domain combined never reached half a percent, before or after.
So when we say ChatGPT cut forum citations by four fifths, what happened in practice is that it stopped citing Reddit. The category label hides a single domain decision. Media/Entertainment feels it most, at 8.6% forum citations against a median sector of 2.6%.
Community and forum in all 7 AI Search Engines
5. Blogs and Personal Sites
3.20% before, 2.24% after, -30%. Independent publishing lost ground quietly alongside journalism. Consumer Goods is the one sector that should care, at 10.2% against a median sector of 3.7%.
6. Video
Video in ChatGPT
0.33% before, 0.98% after, +197%. Tripled, but from a base so small the move is worth less than the percentage suggests.
Video is YouTube, and YouTube tripled
youtube.com is 56.0% of everything ChatGPT cites as video, and it went from 0.150% of all ChatGPT citations before August to 0.510% after 14 August, a rise of 239%. The rest is a long tail of streaming and library domains that moves with it rather than against it.
Video in all 7 AI Search Engines
7. Social Media
Social Media in ChatGPT
0.26% before, 0.84% after, +223%. The largest relative gain in the study, and also one of the smallest categories.
LinkedIn and Facebook, and not TikTok
Two platforms account for nearly all of it. linkedin.com is 42.6% of the category and rose 252%, facebook.com is 19.8% and rose 527%, instagram.com 5.5% and rose 798% off almost nothing.
The absence is the more useful finding. ChatGPT cited TikTok 163 times in two months across every category and every industry. Over the same days Google AI Mode cited it 1,650 times and Google AI Overviews 1,004. X is smaller still, at 21 ChatGPT citations including twitter.com.
If your strategy assumes AI Search reads social platforms the same way, it does not. TikTok is a real source inside Google’s engines and an empty set inside ChatGPT.
Social Media in all AI Search Engines
8. Encyclopedia
Encyclopedia in ChatGPT
1.45% before, 1.43% after, -1%. Flat across both August dates. Pharma is the only sector where it matters, at 4.8% against a median sector of 0.8%.
Wikipedia is not the biggest encyclopedia
wikipedia.org, counting every language edition, is 17.3% of the category, and it rose 47% after 8 August. The other 82.7% is larger and more specialised: pubmed.ncbi.nlm.nih.gov alone outranks Wikipedia, and drugs.com and statista.com both fell by around 60% across the same break.
Encyclopedia in all AI Search Engines
9. Education, and Everything Else
Education went 4.95% before to 4.84% after, -2%. Flat throughout. Whatever changed on 8 August did not touch course and university domains. It is a rounding error in 16 of the 17 sectors and the whole game in the seventeenth: Educational Platforms draws 30.7% of its citations from education domains, against a median sector of 1.1%.
Other went 5.39% before to 4.22% after, -22%. Unclassified and miscellaneous domains shrank as sourcing consolidated. Nothing in it is actionable, which is why it has no chart.
Who Leads Each Category
In 7 of the 17 industries, one brand leads on all seven engines: Coursera in Educational Platforms, Tesla in Energy, YouTube in Media/Entertainment, Pfizer in Pharma, Zillow in Real Estate, Nike in Sports Brands and Uber in Transportation.
The engines are not forming independent opinions. Where a category has a default answer, all seven have converged on the same one. The contested sectors are the opportunity: Consumer Goods, Government, Hospitality and Nonprofit each have four different brands leading on different engines.
ChatGPT Ads + Shopping Cards Per Industry
ChatGPT is the only engine in the study running commercial units, and its two commercial surfaces behave in opposite ways.
Shopping cards track intent exactly. Consumer Goods draws 27.3% and Sports Brands 18.3%. Eight sectors return none at all. Where there is a product to put in a basket, the card appears. Where there is not, it never does.
Ad load does the opposite. Fifteen of the 17 sectors sit inside a narrow band, and commercial and non commercial categories score alike: Real Estate, AI Data Governance and Government all sit within a few points of Consumer Goods and Retail. Only Healthcare and Nonprofit fall clearly below. These figures rank industries against each other and nothing more, because the underlying field is a platform metric rather than a study construct, so it should not be read as the share of prompts that return an ad.
One more gap worth naming. Across 107,633 ad units we identified 6,592 distinct advertisers, and almost none of the largest appear among the most mentioned brands organically. Paid presence and earned presence are two different populations.
How the Engines Differ
Every engine answers the same questions on the same days, so the share of citation volume each one produces is a direct measure of how deeply it sources. Google AI Mode accounts for 21.02% of all citations recorded and Microsoft Copilot 7.78%, a spread of roughly 2.7x between the deepest and the shallowest.
On a simple average ChatGPT looks deepest, but its volume doubled part way through the period, which drags the mean upward for reasons unrelated to its normal behaviour. On the median day Google AI Mode is deepest, and where a distribution has a step change in the middle of it the median is the number to trust.
Concentration splits them in a way that is easy to misread. Google AI Mode reaches half of its citations with just 436 domains, the tightest of any engine, yet it cites 69,390 distinct domains, the widest of any engine. It runs a very hard core and a very long tail at the same time. Claude does the opposite, citing only 19,732 distinct domains in total.
What To Do About It
- Find your row in the grid before you fund anything. Journalism runs from 4.4% to 32.2% and company owned from 29.6% to 77.4%. Those two numbers decide whether your next hire is a PR lead or a content lead.
- Start with the pages you already own. The single largest source of AI answers is a page you can edit today, and for most brands it was written for a machine that ranks rather than one that summarises.
- Look outside marketing if you are in a regulated sector. A quarter of Pharma citations are regulators and 42.4% of Nonprofit citations are government and NGO. No content calendar reaches them.
- Check whether your category is already decided. If one brand leads on all seven engines you are running a displacement campaign, not a visibility campaign.
- Judge your share against your category, not the study. A 5% share of mentions is a strong result in Nonprofit and a failure in Educational Platforms.
- Never diagnose a drop from one engine. ChatGPT changed twice in one month while no other engine did the same. Check a second engine before you rewrite anything.
- Measure category mix, not just your own citations. The 8 August shift was invisible as a brand metric and obvious as a category metric.
Final Conclusion
There is no single AI Search. Journalism dependence varies 7.3x across 17 sectors, and the study average describes almost none of them.
Company owned pages are the largest surface, at 53.7% overall and 77.4% at the top, but only 29.6% in Nonprofit.
Some sectors run on a completely different web. Nonprofit is 42.4% government and NGO, Educational Platforms 30.7% education, Pharma 26.2% regulator.
ChatGPT rewired itself twice in August, halving journalism and cutting forums by four fifths, while no other engine changed level and stayed there.
Depth and concentration separate the engines as sharply as source mix separates the sectors.
Got a GEO experiment idea? Let’s collaborate
At OtterlyAI, we don’t write about AI Search from the sidelines. We test it. With the community, we run and update a public GEO Experimentation Tracker showing what we are running and what moved AI Search. If you are running your own GEO experiment, email rick.tousseyn@otterly.ai and we may feature it with your name attributed.
👉 Want to measure your AI Visibility? Start with OtterlyAI and our GEO Guide.
👉 Learn how to connect your Claude or ChatGPT to OtterlyAI’s MCP
