When AI Engines Cite the Reviewer vs. the Brand: A 25-Vertical Split

The finding

If you ask an AI engine to recommend a product, the structure of the answer depends on what you’re buying. For consumer goods, the engine almost always reaches for a third-party reviewer first. For B2B software, the engine almost always reaches for the vendor’s own domain.

In our Q2 2026 reference dataset, 80.0% of CPG / retail responses cite at least one neutral reviewer (Wirecutter, Sleep Foundation, Outdoor Gear Lab, Healthline, Good Housekeeping, Forbes Advisor, and similar editorial brands). In tech SaaS, the figure is 44.7%. The gap reverses for vendor first-party citations. 92.7% of tech SaaS responses cite the vendor’s own site versus 22.7% for CPG / retail.

Why it matters

In CPG / retail, the path to being recommended runs through a small set of editorial publishers. AI engines treat Sleep Foundation, Outdoor Gear Lab, and Wirecutter as the canonical voice on mattresses, camping gear, and household goods. The brand’s own site is, at best, a supporting citation: 9.3% of CPG / retail responses cite a vendor site without an aggregator cross-reference; 66.7% cite an aggregator without a vendor cross-reference.

In tech SaaS, the structure flips. Engines reach for the vendor’s own domain (pricing pages, product pages, docs, blog content) 49.3% of the time without any neutral reviewer cross-reference. The same response cites an aggregator-only view just 1.3% of the time. For a SaaS buyer, the AI’s answer is essentially the vendor’s own marketing copy, surfaced one layer of abstraction higher.

The strategic consequence: in CPG / retail, getting into AI search means getting into the aggregators. In tech SaaS, the aggregators won’t save you. Your own site has to be the canonical answer.

By category: the three landscapes

Each row shows the share of buyer-intent responses in that category that cited at least one aggregator-family domain, at least one vendor-first-party domain, or only one of the two. “Only” rows are mutually exclusive; they count responses where the engine did not reach for the other side at all.

Consumer services sits in between: telemedicine, online learning, and travel booking pull in both aggregator and vendor voices in roughly equal weight, but mental health apps and real estate platforms pull the category average toward the vendor side. Real estate is structurally unusual. The “aggregators” for real estate are themselves vendor-classified marketplaces (Zillow, Redfin, Realtor.com), so editorial reviewers don’t show up at all.

By vertical: 25-vertical ranking

All 25 verticals, sorted by aggregator share (highest first). For each vertical, n = 15 responses (3 prompts × 5 engines). A vertical at 93.3% aggregator share cited at least one neutral reviewer in 14 of its 15 buyer-intent responses.

Three observations stand out from the long table:

1. The five verticals at the top of the table (running shoes, mattresses, outdoor gear, pet care, supplements, cookware) are not just “CPG.” They are the verticals where a small set of authority publishers (Sleep Foundation, Outdoor Gear Lab, RunRepeat, Healthline, Good Housekeeping) has spent a decade building editorial dominance. AI engines are now consolidating around those same publishers.
2. The five verticals at the bottom (real estate, AI tools, sales engagement, developer infrastructure, product analytics) are dominated by vendor first-party citations with almost no editorial counterweight. Three of those five are tech SaaS subcategories where there is no Sleep Foundation analog. The market never produced one, and the AI engines reflect that.
3. Two verticals (telemedicine and marketing automation) appear high on aggregator share (86.7% and 80%) and on vendor share (93.3% each). Those are the “both” verticals: AI engines consult both the vendor and a reviewer in nearly every response. For a SaaS marketer, those are the verticals where a Wirecutter-style play might still work.

The most vendor-only verticals

These are the verticals where AI engines reach for the vendor’s own site without citing any neutral reviewer. The AI search recommendation is, in effect, the vendor’s own marketing copy re-summarized.

The most aggregator-only verticals

The opposite list: verticals where AI engines almost never cite the brand’s own site. If you sell mattresses, cookware, outdoor gear, or athletic shoes, the AI search recommendation is essentially being written for you by Sleep Foundation, Good Housekeeping, Outdoor Gear Lab, or RunRepeat.

By engine: ChatGPT is the most vendor-trusting

The five engines disagree sharply on how much weight to give the aggregator family. ChatGPT’s answers cite an aggregator in only 26.7% of responses, barely a third of the rate of Google AI Overview (78.7%) or Gemini (77.3%).

Both Google-family engines (AI Overview and Gemini) lean hardest on editorial publishers. This is consistent with Google’s long-standing algorithmic preference for trusted commerce-review sources in traditional search. ChatGPT’s vendor-leaning posture is the outlier and aligns with its product behavior of giving direct, confident answers from primary sources rather than aggregating secondary reviews. Perplexity sits between the two, slightly more vendor-trusting than Gemini, slightly more aggregator-trusting than Claude.

By intent: shortlist queries lean to aggregators

The 75 prompts in the benchmark span three buyer intents: discovery (“best X for Y use case”), shortlist (“X vs. Y vs. Z comparison”), and variation (“cheapest X”, “X for beginners”, “X under $50”). The aggregator-vs-vendor balance shifts as the buyer’s question sharpens.

Discovery and variation queries pull aggregator and vendor citations in rough parity. Shortlist queries are the exception: vendor share drops to 52.0% while aggregator share holds. When a buyer asks for a head-to-head comparison, AI engines reach for the comparison-page editorial sites. This is exactly the moment in the funnel where a vendor’s own marketing is least trusted.

Implications for AI search strategy

Three strategic conclusions follow from the data.

If you sell CPG or retail goods, AI search visibility is bought, mostly, by being in Sleep Foundation, Outdoor Gear Lab, Wirecutter, Healthline, or Forbes Advisor. Your own site rarely appears in the response. Optimizing your product pages for AEO is worth doing, but a content-PR program targeting the small set of category-authority publishers is higher leverage. The exception is skincare / beauty, where vendor citations outpace aggregator citations. That is likely because the major skincare brands run editorial-quality blog content that competes with the niche-publisher hubs.

If you sell tech SaaS, aggregators won’t save you. Most engines will not consult one. Foglift’s own vertical (AI tools / GEO / AI search optimization) sits at 93.3% vendor-only and 6.7% aggregator share, the most vendor-dominated row in the table. For a SaaS buyer the AI search experience is direct vendor citation or nothing. Implication: your own AEO surface (homepage, product pages, docs, comparison pages, blog content) is the entire game. There is no Sleep Foundation for sales engagement platforms or developer infrastructure. You have to be the canonical answer.

If you sell a service in a hybrid vertical(telemedicine, marketing automation, online learning, travel), both levers matter and either one can move the needle. The “both” verticals are the rare cases where a content-PR play and an AEO play stack on top of each other rather than being substitutes.

The 70-point gap between tech SaaS’s vendor-first-party share and CPG / retail’s vendor-first-party share is the single largest structural divide in the Q2 2026 dataset. It implies that the playbook called “AI search optimization” is actually two different playbooks wearing the same name.

Caveats and limits

The binary aggregator-vs-vendor lens is a simplification. Several domains in the “other” group (community UGC like Reddit, video like YouTube, marketplaces like Amazon and Chewy, institutional sources like NIH and .gov) also influence the answer, and their share varies sharply by engine and vertical (the parent benchmark’s per-engine citation-type distribution covers this in detail). The domain taxonomy is hand-curated for the top of the distribution and falls back to heuristics in the long tail; unclassified citations run at roughly 21.6%-31.6% of the “other” bucket. A 15-response sample per vertical is large enough for the headline gap (CPG / retail vs. tech SaaS, 70 points) but small enough that individual-vertical figures should be read with a ±13-point margin of error at 95% confidence. The data was collected on a single day (2026-05-18) and the engines’ underlying training and retrieval indexes drift; this artifact will be refreshed in Q3 2026 with the next benchmark run.

Reproducibility

Every number on this page is derived from the same raw response JSONL used by the Q2 2026 reference benchmark, plus the hand-curated domain taxonomy and a deterministic aggregation script. Identical inputs produce identical outputs.

• Parent reference dataset: AI Search Citation Benchmark: Q2 2026
• Aggregated CSV (domain × engine × vertical × category): /research/citation-benchmark-2026-q2.csv
• Companion analysis (the four-pattern framework): The Foglift AI Citation Map

Want to see whether AI engines cite your own site or your category’s aggregators when buyers ask about you? Run a free Foglift scan. The same engines, the same grounded production models, applied to your URL and your tracked prompts.