The AI Agent Recommendation Index (November 2026)
Which AI agent brands ChatGPT, Gemini, Perplexity and Google AI Mode name for 58 real buyer questions: share of answers per brand and engine, the sources each engine reads, how often answers change day to day, and the raw data.
Across 1,216 answers to 58 buyer questions asked between November 2 and 7, 2026, ChatGPT, Gemini, Perplexity and Google AI Mode named Zapier, Microsoft Copilot Studio, Lindy, n8n and Google Gemini most often: each appeared in 16.7% to 18.9% of answers on average. SPIRITT, which publishes this index and is one of the 45 brands tracked, ranked 17th at 6.9%. That is low, and it is uneven: Perplexity named SPIRITT in 15% of answers, Gemini in 10%, ChatGPT in 2% and Google AI Mode in under 1%.
No single engine agrees with another on the top pick, and the same engine changes its list from one day to the next. The full table, three charts, the method and the raw CSV files are below.
Key findings
- Four brands own the category answer. Zapier (18.9%), Microsoft Copilot Studio (18.3%), Lindy (17.4%) and n8n (16.8%) lead across engines. For the core question "which AI agent should do real work for my company", Copilot Studio was named in 55% of answers, Lindy in 51% and Zapier in 47%.
- Each engine has its own favorites. Gemini names Zapier most (26%). Google AI Mode leans to developer frameworks: CrewAI (21%) and LangGraph or LangChain (18%), both far below that on ChatGPT. ChatGPT names Claude Cowork (13%) and OpenClaw (12%) about twice as often as the other engines.
- Google AI Mode reads Reddit. reddit.com appears in 30% of AI Mode answers and youtube.com in 14%. ChatGPT reads vendor sites and list articles; Perplexity reads the pages written for the exact question.
- Answers are unstable. Asked the same question on the next day, Gemini and Google AI Mode named a different first brand more than half the time. Even the steadiest engines (ChatGPT and Perplexity) kept the same first brand only 62% to 63% of the time.
- SPIRITT is cited more than it is named. ChatGPT used a spiritt.ai page as a source in 10.4% of answers but named SPIRITT in only 2.3%.
The index: AI agent brands named most often
The average gives each engine equal weight. A brand counts as named when the answer text names it; a brand is not counted on a question that already names it (for example, Manus on "What are the best Manus alternatives?").
Share of answers naming each brand, averaged across four AI engines, November 2 to 7, 2026
| Rank | Brand | Average | ChatGPT | Gemini | Perplexity | Google AI Mode |
|---|---|---|---|---|---|---|
| 1 | Zapier | 18.9% | 18% | 26% | 16% | 16% |
| 2 | Microsoft Copilot Studio | 18.3% | 22% | 14% | 22% | 16% |
| 3 | Lindy | 17.4% | 22% | 23% | 18% | 7% |
| 4 | n8n | 16.8% | 20% | 21% | 18% | 9% |
| 5 | Google Gemini | 16.7% | 19% | 12% | 21% | 16% |
| 6 | Manus | 13.1% | 15% | 13% | 17% | 7% |
| 7 | Claude Code | 12.1% | 12% | 13% | 11% | 12% |
| 8 | CrewAI | 11.2% | 8% | 9% | 7% | 21% |
| 9 | LangGraph / LangChain | 11.2% | 6% | 15% | 6% | 18% |
| 10 | Cursor | 10.8% | 8% | 14% | 12% | 10% |
| 11 | Gumloop | 9.6% | 8% | 14% | 10% | 7% |
| 12 | Salesforce Agentforce | 9.3% | 10% | 9% | 12% | 6% |
| 13 | Make | 8.6% | 7% | 10% | 10% | 7% |
| 14 | Relevance AI | 8.3% | 11% | 7% | 8% | 7% |
| 15 | Replit | 7.8% | 9% | 8% | 9% | 5% |
| 16 | Claude Cowork | 7.1% | 13% | 6% | 6% | 3% |
| 17 | SPIRITT | 6.9% | 2% | 10% | 15% | <1% |
| 18 | OpenClaw | 6.7% | 12% | 5% | 6% | 3% |
| 19 | Bolt | 6.3% | 7% | 6% | 5% | 7% |
| 20 | OpenAI Codex | 5.4% | 8% | 4% | 7% | 2% |
The full list of 45 brands, with share of voice and how often each brand is named first, is in the index CSV.
Each engine has its own favorites
Share of answers naming each brand, per engine, November 2 to 7, 2026
- ChatGPT names the most brands per answer (3.2 on average) and reads the most sources (15 per answer). Lindy is its most frequent first pick (8.7% of answers start with it), then Claude Cowork and Copilot Studio.
- Gemini is the most Zapier-heavy engine and, with Google AI Mode, the most volatile. It names few sources (3.3 per answer), so one article can shape a whole answer.
- Perplexity is the steadiest engine and the one that reads a question's dedicated pages most directly. It is also where SPIRITT does best, for reasons explained below.
- Google AI Mode names the fewest brands (2.4 per answer) and favors frameworks developers build with (CrewAI, LangGraph) over finished products.
Which brands win each buyer decision
| Buyer decision | Answers | Most named brands | SPIRITT |
|---|---|---|---|
| D1. Which autonomous AI agent should do real work for my company? | 167 | Microsoft Copilot Studio 55%, Lindy 51%, Zapier 47% | 5% |
| D2. What should I use instead of the tool I know? | 251 | Lindy 36%, Manus 31%, n8n 30% | 3% |
| D3. Which AI can build and ship real software to production? | 147 | Replit 41%, Lovable 37%, Bolt 35% | 8% |
| D4. Which AI can run a growth or sales job end to end? | 168 | 11x 21%, Artisan 20%, Clay 20% | 6% |
| D5. Which AI can run my operations and recurring workflows 24/7? | 147 | Lindy 26%, Zapier 18%, Google Gemini 18% | 13% |
| D6. Which agent has the capabilities I need? | 189 | Claude Code 19%, LangGraph / LangChain 18%, CrewAI 18% | 13% |
| D7. Which AI agent platform is safe for my enterprise? | 84 | Amazon Bedrock AgentCore 24%, n8n 21%, Salesforce Agentforce 19% | 7% |
| D8. What does it cost and is it worth it? | 63 | Zapier 14%, SPIRITT 14%, Clay 8% | 14% |
Sales questions are the most fragmented: no brand reaches a quarter of answers, because engines split them across AI SDRs (11x, Artisan), data tools (Clay) and calling tools.
Where SPIRITT stands
SPIRITT is a tracked brand in its own index, so here is its result without rounding up.
| Engine | Answers | SPIRITT named | spiritt.ai page cited | SPIRITT rank (of brands named) | Named first |
|---|---|---|---|---|---|
| ChatGPT | 346 | 2.3% | 10.4% | 40 of 45 | 2.0% |
| Gemini | 348 | 9.8% | 11.2% | 12 of 43 | 7.5% |
| Perplexity | 348 | 14.9% | 30.8% | 7 of 45 | 12.4% |
| Google AI Mode | 174 | 0.6% | 0.6% | 40 of 43 | 0.6% |
Three things explain the pattern:
- The rate rose during the week because we published answers. On November 2, no engine named SPIRITT. SPIRITT published guides written for these exact questions during the same week, and by November 7 the daily rate was 24% on Perplexity, 19% on Gemini and 7% on ChatGPT. Most SPIRITT mentions come from engines reading those spiritt.ai guides, not from independent reviews. Third-party coverage of SPIRITT is still thin.
- ChatGPT reads SPIRITT pages but rarely names SPIRITT. In about 8 of every 100 ChatGPT answers, a spiritt.ai page was among the sources and the answer still recommended other brands. ChatGPT mostly names brands that its other sources (vendor sites, directories, list articles) already agree on.
- Google AI Mode has not picked SPIRITT up at all. It leans on Reddit threads and YouTube videos, where SPIRITT has almost no presence yet.
The question set also leans toward SPIRITT's strengths (own cloud computer, every model in one place, pay-per-use data), because these are the questions SPIRITT wants to answer well. That bias favors SPIRITT, and its rate is still low.
The sources each engine reads
| Engine | Most-read domains (share of answers that cite it) | Sources per answer |
|---|---|---|
| ChatGPT | zapier.com 24%, lindy.ai 19%, logospng.org 18%, manus.im 16%, microsoft.com 15%, n8n.io 15%, dupple.com 15%, resources.rework.com 14% | 15.0 |
| Gemini | vellum.ai 17%, spiritt.ai 11%, lindy.ai 10%, simular.ai 6%, manus.im 5%, zapier.com 4%, jetadmin.io 4%, surething.io 3% | 3.3 |
| Perplexity | spiritt.ai 31%, taskade.com 18%, sim.ai 14%, github.com 13%, resources.rework.com 12%, hostinger.com 12%, zapier.com 11%, gumloop.com 10% | 9.1 |
| Google AI Mode | reddit.com 30%, youtube.com 14%, github.com 7%, zapier.com 5%, lindy.ai 5%, medium.com 4%, sim.ai 4%, openai.com 3% | 2.5 |
Only four domains are in the top 25 of all four engines: github.com, lindy.ai, manus.im and zapier.com. Two patterns stand out. ChatGPT cites logo download sites (logospng.org) in 18% of answers, likely when it adds brand images to a list. And many of the most-read pages are vendor blogs that rank their own product first (Vellum, Taskade, Sim, Gumloop, Lindy and SPIRITT all appear), so a large share of what AI engines repeat comes from vendors.
How often AI answers change day to day
Same question, same engine, consecutive days: how much of the answer stays the same
For every question and engine we compared each day's answer with the next day's (984 pairs).
| Engine | Same first-named brand | Overlap of brands named | Identical brand list | Overlap of sources |
|---|---|---|---|---|
| ChatGPT | 63% | 65% | 33% | 24% |
| Gemini | 44% | 45% | 19% | 23% |
| Perplexity | 62% | 67% | 38% | 56% |
| Google AI Mode | 39% | 38% | 17% | 24% |
Overlap is the Jaccard index: shared items divided by all items across both days. ChatGPT keeps roughly the same brands while only about a quarter of its sources overlap from day to day, so its brand picks are more settled than its reading list. Gemini and Google AI Mode change both.
What this means for buyers
- Ask more than one engine. The engines agree on four or five household names and disagree on the rest. A brand missing from one answer is often the first pick in another.
- Ask twice. With a different first pick on the next day in 37% to 61% of cases, a single answer is a sample, not a verdict.
- Check who wrote the source. Many cited pages are vendor blogs ranking their own product, SPIRITT's included. Open the sources and weigh independent tests higher.
- Match the brand to the job. Frequency reflects how much has been written about a brand, not how well it fits your task. Workflow builders (Zapier, n8n, Make), agent frameworks (CrewAI, LangGraph), coding tools (Cursor, Claude Code) and autonomous workers (Manus, Lindy, SPIRITT) solve different problems, even when one answer lists them together.
- Test on one real task. Give two or three finalists the same piece of real work and compare the finished result. For a hands-on comparison of agents on the same tasks, see the general purpose AI agents task test and the best AI agents in 2026.
Method
Engines and collection. ChatGPT, Gemini and Google AI Mode answers were collected from their consumer interfaces with a US location through a data API. Perplexity answers came through a data API until November 6 and through Perplexity's own API with web search from November 7. Each engine got the question as written, once per day, with no follow-up questions. Google AI Overviews were also collected but answered under half of the keyword searches, so they are left out.
Dates. November 2 to 7, 2026 (six days) for ChatGPT, Gemini and Perplexity; November 5 to 7 for Google AI Mode, when our collection for it began. Engine-days with fewer than half the questions answered are excluded. Total: 1,216 answers.
Questions. 58 buyer questions in 8 decisions, none of which name SPIRITT. They are the questions the SPIRITT team tracks every day, so they cover autonomous agents broadly and lean toward capabilities SPIRITT offers. Every question is in the answers CSV; a sample per decision:
- D1 (8 questions): "What is the best autonomous AI agent for a small business that can actually do the work, not just chat?", "What are the best general purpose AI agents right now?"
- D2 (12): "What are the best Manus alternatives?", "What are the best Zapier alternatives that use AI agents instead of zaps?"
- D3 (7): "Which AI can build a full-stack web app and deploy it to production for me?"
- D4 (8): "What is the best AI SDR that does cold outreach end to end?"
- D5 (7): "Which AI agent runs 24/7 in the cloud even when my laptop is closed?"
- D6 (9): "Which AI agents have their own cloud computer and can use a browser?"
- D7 (4): "What are the most secure AI agent platforms for enterprises?"
- D8 (3): "How much does an autonomous AI agent cost compared to hiring an employee?", "Is Manus worth it in 2026?"
How mentions were detected. We matched 45 brand names in each answer's text and in the named entities the engine returned, with case-sensitive rules so that common words do not count ("make sure" is not Make, "the operator" is not OpenAI Operator, a "cursor" is not Cursor). Product families are grouped: "ChatGPT agent / Operator", "Microsoft Copilot Studio" (also Microsoft 365 Copilot), "Perplexity Comet / Computer", "LangGraph / LangChain". A brand is not counted on a question that already names it.
Metrics. Mention rate: share of an engine's answers that name the brand. Average: mean of the four engine rates. Share of voice: the brand's mentions over all brand mentions on that engine. Named first: the brand is the first tracked brand in the answer. Cited: a page on the brand's own domain is among the answer's sources.
Limits. A mention is not always a recommendation: an answer can name Zapier as an integration or Manus as the tool to replace. Brands outside our 45-name list are not counted, so the index under-counts niche tools (the CSV keeps every source domain for your own analysis). "Google Gemini" mentions include Gemini naming itself. Six days is a short window, and SPIRITT's own publishing changed its numbers during it. Perplexity's collection method changed on November 7, so its source counts before and after are not strictly comparable. Source counts differ partly because each engine exposes a different number of sources.
Download the data
- Index CSV: one row per engine and brand (176 rows): answers, mentions, mention rate, share of voice, named-first rate, overall rank.
- Answers CSV: one row per answer (1,216 rows): date, engine, question, brands named in order, first brand, whether SPIRITT was named or cited, and every source domain.
Both files are free to use under CC BY 4.0 with a link to this page. We plan to publish this index every month with the same questions and method.
FAQ
Which AI agent do ChatGPT, Gemini and Perplexity recommend most?
In November 2026, across 58 buyer questions, the most named AI agent brands were Zapier (18.9% of answers on average), Microsoft Copilot Studio (18.3%), Lindy (17.4%), n8n (16.8%) and Google Gemini (16.7%). The engines differ: Gemini names Zapier in 26% of answers, ChatGPT and Perplexity name Copilot Studio in 22%, and Google AI Mode names CrewAI in 21%.
Where does SPIRITT rank in the AI Agent Recommendation Index?
SPIRITT ranked 17th of 45 brands with an average of 6.9%: 14.9% of Perplexity answers, 9.8% of Gemini, 2.3% of ChatGPT and 0.6% of Google AI Mode. SPIRITT publishes the index, and most of its mentions came from engines reading spiritt.ai guides published that week.
What sources do AI engines use to recommend AI agents?
ChatGPT reads vendor sites (zapier.com, lindy.ai, manus.im, microsoft.com), list articles and logo sites. Perplexity reads pages written for the exact question. Google AI Mode cites reddit.com in 30% of answers and youtube.com in 14%. Only github.com, lindy.ai, manus.im and zapier.com are in the top 25 sources of all four engines.
How often do AI answers about AI agents change?
Often. Asked the same question on the next day, ChatGPT and Perplexity kept the same first-named brand about 62% to 63% of the time, Gemini 44% and Google AI Mode 39%. Only 17% to 38% of next-day answers named exactly the same set of brands.
Can I use this data?
Yes. Both CSV files are free under CC BY 4.0 with a link to this page. The answers file has one row per answer with the question, the brands named in order and every source domain.
How was the AI Agent Recommendation Index measured?
Each of 58 buyer questions was asked once a day in the US to ChatGPT, Gemini and Perplexity (November 2 to 7, 2026) and Google AI Mode (November 5 to 7), 1,216 answers in total. Brands were matched by case-sensitive name rules in the answer text and named entities, and a brand was not counted on a question that already names it.
How we checked
SPIRITT publishes this index, and SPIRITT is one of the brands it tracks. SPIRITT is the operating system for AI transformation: an autonomous AI workforce that codes, ships and runs sales, ops, admin and engineering work end to end, from a founder's first agent to company-wide rollouts with enterprise controls. The SPIRITT agent ran the collection and analysis for this index in a SPIRITT workspace. Every number on this page comes from the attached CSV files; no answer was edited or removed.
