AI Engines: how each one cites, who uses them, and what to optimize for
Most people use “AI search” as a single category. It’s not. ChatGPT and Perplexity look like they do the same thing but cite very different sources for the same query. Microsoft Copilot uses Bing’s index just like ChatGPT browse does and still produces a different output. Grok pulls from X posts in ways no other engine does. Meta AI cites Instagram captions on consumer queries.
Different engines, different users, different optimization moves.
This is the hub for everything we’ve published on how each AI engine actually works, who uses it, what gets cited, and what to do about it. If you’re trying to figure out which engines to prioritize for your buyer, start here.
The engines that matter for B2B in 2026
Five major engines plus three smaller ones worth watching.
The big five are ChatGPT (browse mode + standalone), Perplexity, Google AI Overviews, Gemini, and Microsoft Copilot. Together they account for the vast majority of AI-driven research traffic in B2B markets.
The three to watch are Grok (X-integrated, niche but growing), DeepSeek (developer-heavy, international), and Meta AI (consumer-skewing, embedded in Instagram and WhatsApp).
We test all eight separately for client engagements. The optimization moves differ across them, and the buyer demographics differ enough that you can’t aggregate.
Start here if you’re new to AI search
If you’ve never thought systematically about which AI engines your buyers actually use, start with Whose AI search? ChatGPT vs Perplexity vs Gemini, sorted by who’s actually using them. It covers the decision framework for picking which engines to prioritize.
Then read The retrieval pipeline behind every AI engine to understand what all these engines have structurally in common. Knowing the shared retrieval pattern makes the per-engine differences more navigable.
Then come back here for the engine-specific deep coverage.
ChatGPT (browse mode + memory)
ChatGPT has the largest active user base of any AI engine. About 600 million weekly active users as of mid-2026. The demographic is broad: consumer, prosumer, professional, B2B buyer.
For B2B citation work, ChatGPT browse mode is the highest-volume channel. The optimization moves are the most-mature, the buyer demographic is the broadest, and the citation visibility per source is high.
Read:
– How ChatGPT actually decides what to cite (three signals tested)
– ChatGPT browse mode vs memory: citation behavior compared
Perplexity
Perplexity has a smaller user base but the user behavior is different. Perplexity users are more likely to be doing serious research with verification expectations. The product was built for citation-first answers, and the user base has self-selected for people who care about source quality.
For B2B brands selling into senior decision makers, analysts, journalists, or high-value buyers, Perplexity often delivers higher-quality citation traffic than ChatGPT despite the lower volume.
Read:
– Perplexity citations: how to actually get included
Google AI Overviews
Google AI Overviews is the AI summary at the top of regular search results. It reaches every Google user who triggers an Overview, which is now a meaningful fraction of all commercial searches.
The optimization for AI Overviews is mostly classic SEO with structured data additions. If your rank in Google’s top 5 organic results, you’re a candidate for AI Overview citation. If you don’t rank, you’re not.
We’ve written about the broader landscape in How to get cited inside ChatGPT, Perplexity, and Google AI Overviews in 2026.
Gemini
Gemini is Google’s AI assistant. It’s integrated across Google Workspace, available standalone, and embedded in Pixel and Chrome. The user base overlaps heavily with Google Workspace users doing professional research inside their daily workflow.
The optimization moves overlap with classic Google SEO because Gemini’s retrieval uses Google’s index. Knowledge Graph presence matters more than for other engines.
Read:
– Gemini cites differently: the Google Maps signal we did not expect
Microsoft Copilot
Microsoft Copilot is the AI assistant in Microsoft 365 plus a standalone consumer version. The user base for B2B citation work is the enterprise buyer in a Microsoft 365 tenant. They use Copilot inside Outlook, Teams, and Word to research vendors.
Same Bing index as ChatGPT browse, but a different ranking layer on top produces different citations. Read:
– Microsoft Copilot citations vs ChatGPT (same Bing index, different results)
Claude
Claude is Anthropic’s AI assistant. It’s used heavily in technical and analytical workflows. The citation behavior is less studied publicly than ChatGPT or Perplexity, but the user demographic skews toward engineering, research, and serious analytical work.
Read:
– Claude AI citations: the dark matter of AI search
Grok
Grok is xAI’s AI assistant, integrated into the X platform. The user base is smaller than the big engines but the citation behavior is distinct: Grok pulls from X posts directly in ways no other engine does. For brands with substantive X presence (especially in tech, crypto, operator culture), Grok matters more than its raw user count suggests.
Read:
– Grok citations: what we tested, what we found, and what we still don’t know
DeepSeek
DeepSeek went from unknown to top-5 AI engine in under 18 months. The user base is global with strong representation in developer communities and regions where Western AI access is limited.
For brands with technical product DNA, GitHub presence, or international content investment, DeepSeek delivers citation traffic the Western engines underweight.
Read:
– How DeepSeek cites: what we know about the engine that came out of nowhere
Meta AI
Meta AI is embedded inside Facebook, Instagram, WhatsApp, and Messenger. The user base is large and skews consumer. For B2B brands, Meta AI is mostly de-prioritizable. For consumer brands and local services, it’s essential.
Read:
– Meta AI search: how it fetches, what it cites, and who should care
The patterns across all engines
Three things hold across every engine we’ve tested.
Every engine that uses retrieval-augmented generation (which is most of them) preferentially extracts from pages with clean structured data. Schema markup, answer-first writing, named-source density. These don’t differ much across engines even though the retrieval layers do.
Every engine weights entity binding to some degree. Knowing who you are, what you do, and what you’re authoritative on requires consistent signal across Wikidata, sameAs schema, Knowledge Graph presence, and cross-platform consistency. The brand without entity binding gets de-weighted across all engines.
Every engine produces wrong citations sometimes. Schema that contradicts visible page content produces wrong citations. Outdated dateModified produces wrong citations. Invented aggregateRating produces wrong citations. The brand’s job is making the structured data and the visible content match.
We’ve also written about the broader category problem in When AI engines refuse to cite you (the 4 disqualifiers), which covers the structural reasons a brand might get filtered out before even entering the candidate pool for any engine.
What we do at NetPageTwo
We track citation rate across all eight engines separately for every client. The first audit produces a baseline number for each. We weight optimization effort by the client’s specific buyer demographic, not by raw engine size.
A typical B2B SaaS client gets weighting like 40% ChatGPT, 25% Perplexity, 15% Google AI Overviews, 10% Gemini, 10% Copilot. An enterprise software client selling into Microsoft 365 shops shifts toward Copilot. A developer tools company shifts toward DeepSeek and ChatGPT. A consumer brand adds Meta AI.
The work order is dictated by the buyer, not by the engine.
If you want a tested baseline of how each engine currently cites your brand, the fit call covers it. The output is a one-page citation report with per-engine recommendations and a 30-day priority plan.
Related hubs:
– Schema for AI search
– AI content strategy
– GEO 101
– The new search
– Field Notes