Where do the AI engines cite from when they recommend a video tool?

Of the 284 source citations across the recorded answers, roughly one in three (94, about 33 percent) point to a video tool vendor's own website. Synthesia, InVideo, and HeyGen sit in the top six most-cited domains alongside Reddit and Zapier. That is a distinct fingerprint: in the companion B2B GTM study, only about 14 percent of citations were vendor first-party. Reddit is the single most-cited domain but accounts for only about 9 percent of all citations, spread across 129 domains.

How is this AI video leaderboard measured?

Each row is a count of recorded live-engine outputs. We submitted 16 buying-intent AI-video prompts to ChatGPT, Perplexity, Gemini, Grok, and Google AI Overviews and read the rendered answer for the tools it named and the domains it cited, for 44 recorded answers. We report each engine separately and never average them. Query coverage is uneven: only 4 of the 16 prompts were answered by all five engines, so we report cross-engine consensus as the number of engines that named each tool, not a single agreement coefficient.

Which AI Video Generator Do AI Engines Recommend Most? (2026 Leaderboard), Nesyona

Name: Which AI video generators do AI engines recommend? Cross-engine naming-frequency and citation dataset (June 2026)
Creator: Vincent Wesley Couey
Published: 2026-06-17
License: https://creativecommons.org/licenses/by/4.0/

The leaderboard: what the engines actually name

Every "best AI video generator" list is somebody's opinion, usually a vendor ranking itself first. This is not that. Each row is a mechanical count of recorded live-engine answers. Tools are ranked by how many of the five engines named them, the cross-engine consensus signal, then by total naming count. The dots show engine breadth; the bar shows raw volume. Click any tool for the per-engine receipts.

Tool

Engines naming

Total mentions

Creatify leads on consensus, not volume: six mentions, but the only tool every engine named. InVideo (14), HeyGen and Synthesia (13 each) win on raw count while missing one engine apiece. A tool named by zero clean engines is effectively invisible to anyone asking an AI which video tool to buy.

Why consensus breadth, and not one agreement number

The companion GTM study reports a single Fleiss kappa because every model answered every category. Here it would be dishonest. Query coverage is uneven: of the 16 prompts, only 4 were answered by all five engines. A cross-engine agreement coefficient computed over asymmetric coverage would be a statistic about which engine happened to answer which prompt, not about how much the engines agree. So we report the honest thing the data supports: how many engines named each tool. Creatify at five of five is a real, unambiguous consensus signal; a tool at one of five is not.

One tool cleared all five engines. Six more (InVideo, HeyGen, Synthesia, Google Veo, CapCut, Higgsfield) cleared four. Below that the field fragments fast: AI video is a young, crowded category with no settled incumbent, so the engines scatter far more than they do in mature software categories.

The citation autopsy: where the engines get their answers

The second layer, and the one a vendor should read twice. We logged every source domain the engines cited: 284 citations across 129 distinct domains. Reddit is the single most-cited domain, but at 25 hits it is only about 9 percent of all citations. The real story is what sits right behind it.

Vendor's own siteForum / video UGCThird-party editorial

One in three citations is a vendor citing territory it owns. Of the 284 citations, 94 (33 percent) point to an AI video tool's own website. Synthesia, InVideo, and HeyGen each land in the top six most-cited domains, ahead of most independent roundups. When an engine recommends one of these tools, it often backs the claim with the tool's own marketing page.

33%

Vendor first-party (the tool's own domain), 94 of 284

13%

Forum and video UGC (Reddit 25, YouTube 13), 38 of 284

54%

Third-party editorial (roundups, review blogs, aggregators), 152 of 284

This is a different fingerprint from B2B software. In the companion AI Citation Autopsy for GTM tools, only about 14 percent of citations were the vendor citing itself; roughly four in five pointed to a third party. AI video inverts a big slice of that. The category is newer and thinner on independent review coverage, so the engines fall back on vendor pages far more often. For a video vendor that means your own site is not just necessary, it is a source the engines are unusually willing to quote, and structured, dated, extractable pages punch above their weight here.

A caveat we state rather than bury: Gemini returns ranked answers with sparse, hard-to-extract source grounding, so no citations were captured from it. The autopsy therefore rests on Perplexity (112 citations), Google AI Overviews (86), Grok (49), and ChatGPT (37). Full domain table in data.json.

The invisible vendors

The most actionable column is the one that is mostly zeros. Several real, shipping AI video tools were named by one engine, once, across all 44 answers.

Pika, Visla, MagicHour, and AdStellar AI surfaced a single time, on a single engine each. Colossyan, Luma Dream Machine, Pictory, PixVerse, and WaveSpeed are close behind, named by at most two engines. When a buyer asks an AI engine which video tool to use, these names do not come up. That is not a quality verdict, several make capable products. It is a visibility verdict, and visibility inside AI answers is now a distribution channel a vendor either owns or forfeits.

Get your verified badge

If an AI engine named your tool in this recorded sample, you earned a Verified by Lattice badge. It is free, and it states only your true recorded standing, how many of the five engines named you, dated. It is not a paid placement and we do not gate it. A tool named by fewer than three of the five engines is not eligible yet; it is on the invisible-vendors list above instead. Pick your tool to preview the badge and copy the one-line embed. The link back is dofollow.

Select your tool

Copy-paste embed (dofollow)

The badge states exactly what the recorded data shows for your tool and nothing more. Full gallery and light or dark variants at nesyona.com/badges. Want a labelled featured spot on this leaderboard? That is the one paid tier, self-serve, always marked Sponsored, and it never changes the counts above: claim a spot.

Reproduce this

Every number on this page is generated from the raw recorded answers. Take the data, re-run the count, and check it yourself.

Leaderboard data JSON Flat table CSV

Method. On 2026-06-17 we submitted 16 buying-intent AI-video prompts (head queries, use-case, constraint, comparison, and "what does Reddit recommend" forms) to five AI engines through their live web interfaces: ChatGPT, Perplexity, Gemini, Grok, and Google AI Overviews. Once each answer finished rendering we read it for the AI video tools it named and the source domains it cited. That produced 44 recorded answers (Perplexity 15, Grok 10, ChatGPT 8, Google AI Overviews 6, Gemini 5) and 284 source citations across 129 domains. Tool-name variants are normalized to one brand (for example "Veo 3.1" and "Google Veo 3.1" both count as Google Veo); a citation to a recognized vendor domain also counts as that engine naming the tool. Engines are reported per engine and never averaged into a blended score. The aggregation is reproducible from the raw capture files with the open measurement engine share-of-voice.mjs#runFromCaptures. Methodology is shared with the Lattice cross-engine program at lucreya.com.

Consensus, not a coefficient. Query coverage is asymmetric: only 4 of the 16 prompts were answered by all five engines, so we deliberately do not report a cross-engine agreement coefficient (a Fleiss kappa over uneven coverage would measure sampling, not agreement). We report the number of engines that named each tool, which the data does support.

Read before citing. This is an exploratory June-2026 snapshot, and four limits bind any claim from it:

Small sample, mostly single runs. Most prompts were run once per engine; AI answers are stochastic, so these are point estimates, not stable rates.
Asymmetric coverage. Engines answered different numbers of prompts (Perplexity 15, Gemini 5), so raw totals partly reflect how often each engine was captured, which is why consensus breadth, not one blended score, is the headline.
Gemini citations not captured. Gemini gives ranked answers with sparse source grounding, so the citation autopsy rests on the other four engines.
Personalization is present, and we floor for it. We do not count Nesyona itself as AI-recommended: nesyona.com appeared only in a logged-in ChatGPT session with prior site history, never in a clean logged-out engine, so it is excluded from every count on this page.

Open data, CC-BY 4.0. Free to share and adapt with attribution.

Couey, V. W. (2026). Which AI Video Generator Do AI Engines Recommend Most? (2026 Leaderboard). Nesyona Research, Lattice network. https://nesyona.com/ai-video-leaderboard

Machine-readable: data.json · data.csv · License CC-BY 4.0. DOI 10.5281/zenodo.21272385 for this dataset.