OtterlyAI’s AI Search Index, updated September 8, analyzed 6,095,520 AI citation events across seven AI-search experiences and 17 U.S. industries. The headline average is simple: company-owned pages supplied 53.7% of citations, news and media 16.9%, and government or NGO sources 8.1%.
The average is also the least useful number in the report. Industrial/Manufacturing drew 77.4% of citations from company-owned sites, while Nonprofit drew 29.6%. Technology drew 32.2% from news and media; AI Data Governance drew 4.4%. The same source strategy cannot describe both ends of those ranges.
That extends a pattern in our earlier reporting on cross-engine citation overlap: AI search is not one distribution surface. The new evidence adds a second segmentation layer. B2B teams need to benchmark source mix by both industry and engine before deciding whether owned content, earned media, institutional sources, communities or another channel deserves the next unit of effort.
Direct answer — Do AI engines cite the same source types across industries?
No. OtterlyAI’s July-August 2026 dataset records 6,095,520 citations across seven AI-search experiences and 17 U.S. industries, with large differences by sector and engine. Company-owned citation share ranges from 77.4% in Industrial/Manufacturing to 29.6% in Nonprofit. That is a 47.8-percentage-point spread, an IVRIS calculation from the published dataset, not evidence of traffic or conversion differences.
- Company-owned pages account for 53.7% of all recorded citations; news and media account for 16.9%.
- Owned-source share spans 47.8 percentage points between Industrial/Manufacturing and Nonprofit. IVRIS calculation based on OtterlyAI’s published dataset.
- Google AI Mode cited 69,390 distinct domains versus Claude’s 19,732, a 3.52x breadth ratio. IVRIS calculation based on OtterlyAI’s published dataset.
- OtterlyAI separately recorded 107,633 ChatGPT ad units and 15,363 shopping cards; those commercial observations are not part of the organic citation-share denominator.
What the 6.1 million-citation dataset actually measured
OtterlyAI says it ran the same thousands of buyer-style questions every day through July and August 2026 across ChatGPT, Google AI Overviews, Google AI Mode, Perplexity, Gemini, Microsoft Copilot and Claude. The prompts covered ranked lists, comparisons, pricing, how-to questions and trust checks. The geography was U.S. only.
A citation means an owned or third-party URL referenced in an AI-generated answer. It does not mean a unique URL, unique domain, visit or recommendation. OtterlyAI reports 6,095,520 citations on “clean collection days,” but does not disclose the exact prompt count, the exact clean-day date bounds, duplicate-handling rules, or whether branded and unbranded prompts were mixed in this Index.
The ten source classes are company owned; news and media; government and NGO; community and forum; blogs and personal sites; other; education; encyclopedia; social media; and video. The Index does not separately disclose marketplace or directory categories.
The industry average hides the useful source-mix gaps
The most actionable finding is dispersion, not the 53.7% overall owned-site share. Industrial/Manufacturing’s 77.4% owned share is 2.61 times Nonprofit’s 29.6%, with a 47.8-point spread. Educational Platforms gets 30.7% of citations from education domains against a 1.1% sector median: a 29.6-point premium, or 27.91 times the median. Both are IVRIS calculations based on OtterlyAI’s published dataset.
Other categories become material only in certain sectors. Nonprofit gets 42.4% from government and NGO sources; Pharma gets 26.2%. Media/Entertainment gets 8.6% from forums against a 2.6% sector median.
One brand leads all seven engines in seven of 17 industries: 41.2% of sectors, leaving 58.8% without complete convergence. IVRIS calculation based on OtterlyAI’s published dataset.
Engine behavior changes the source strategy again
Industry is only half the problem. Google AI Mode accounted for 21.02% of all citations recorded, versus 7.78% for Microsoft Copilot. It also cited 69,390 distinct domains across the period, compared with Claude’s 19,732. The difference is 49,658 domains, and AI Mode’s distinct-domain breadth is 3.52 times Claude’s. IVRIS calculation based on OtterlyAI’s published dataset.
ChatGPT cited TikTok 163 times during the period; Google AI Mode cited it 1,650 times, a 10.12x difference. IVRIS calculation based on OtterlyAI’s published dataset.
August also shows why engine-specific baselines need dates. OtterlyAI reports ChatGPT’s news-and-media share falling from 21.33% in its baseline table to 10.40% after August 14, while Perplexity moved from 22.7% to 23.4%. The vendor associates the ChatGPT break with a model rollout, but the safe inference is narrower: the observed ChatGPT source distribution changed while the other tracked engines did not make the same sustained move.
What B2B teams should use these statistics for
Our read: use this dataset as a source-allocation benchmark, not a ranking recipe. Start with the closest industry row, then compare the engines your buyers use before deciding whether the next investment belongs in owned evidence, earned media, institutional sources, communities, social or video.
Keep the measurement boundary intact. Citation share is not traffic share, click share, ranking probability, conversion rate, recommendation probability or market share. A source can be cited without the brand being named at all, as our analysis of ghost citations shows.
There are also unresolved denominator questions. The Index describes the study as July-August 2026, while its ChatGPT before/after table labels the baseline “Jun – Aug.” The precise clean-day inclusion rule is not published. OtterlyAI’s broader research methodology says AI-search findings apply to the tracked prompts, platforms and time period rather than proving universal platform mechanisms.
For your own reporting, keep these benchmarks separate from first-party evidence. Our AI search visibility measurement framework treats citations, mentions, referrals and business outcomes as different measurements.
Frequently Asked Questions
OtterlyAI reports 6,095,520 citations on clean collection days across seven AI-search experiences and 17 U.S. industries. The report describes these as citations generated from the same thousands of buyer-style questions run daily through July and August 2026; it does not identify them as 6,095,520 unique URLs or domains.
Across OtterlyAI’s full dataset, company-owned pages have the largest citation share at 53.7%, followed by news and media at 16.9% and government or NGO sources at 8.1%. Those overall shares hide large sector differences, so the study-wide average should not be treated as a universal target mix.
No. Citation share describes how references in the observed AI answers were distributed across source categories. It does not measure clicks, visits, recommendation probability, conversion rate, revenue or market share. Those outcomes require separate first-party or platform-specific evidence and should not be inferred from this citation dataset.
The OtterlyAI dataset argues against treating one source mix as universal. Source-category shares vary by industry, and engines also differ in citation volume, domain breadth and their use of sources such as journalism or TikTok. A defensible plan therefore needs an industry baseline and an engine-specific baseline before tactics are chosen.
