Do AI Engines Agree on the Best Software? The AI Search Disagreement Index (2026 Data)

How often do AI engines agree on the best tool?

Cross-engine agreement is the rate at which independent AI engines name the same top tool for one buying question, and in the current snapshot a clean five-engine match happened in zero of sixteen categories.

The headline number is 0 percent, and it surprised us too. We ran 16 B2B software categories through 5 AI engines and logged the tool each engine named first. Not once did all five land on the same single pick. The audit captured 716 recommendations naming 242 distinct tools, and more than half of those tools (128 of 242) were named by exactly one engine. The market is wide, each engine reads a different slice of it, and the result is that "the best CRM" or "the best email tool" depends heavily on which chatbot the buyer opened.

Source: The AI Recommendation Audit (2026), 5 engines × 16 categories. Snapshot 2026-06-19. DOI 10.5281/zenodo.20767878.verified 2026-06-19

The "best tool" the AI gives you is mostly the engine you happened to ask, not a settled fact.Bottom line

Why two different agreement numbers, 0 percent and 36 percent?

The index draws on two dated runs with different scope, and reporting both honestly is the point: a tighter three-engine GTM run agreed 36 percent of the time, while the broader five-engine B2B run agreed 0 percent.

The number moves with how many engines and how broad a market you test. Our first run asked 20 go-to-market buying questions to three browser engines (ChatGPT, Perplexity, Google AI Overviews) and got full agreement on 36 percent of the category and intent queries. The newer run widened to 16 general B2B software categories and 5 engines, including model-memory engines (Llama via Groq, Cohere) that answer from training rather than live search, and agreement collapsed to 0 percent. More engines and a broader market mean more room to disagree. Both figures are real, both are dated, and the index shows both rather than picking the flattering one.

Two snapshots, different scope. Sources: Who AI Recommends: GTM (2026) (DOI 10.5281/zenodo.20632768) and The AI Recommendation Audit (2026) (DOI 10.5281/zenodo.20767878).verified 2026-06-19

The disagreement index: where engines split and where they converge

Agreement is not random; it tracks how consolidated a category is, so the index sorts categories by whether a dominant winner exists.

The engines agree where one tool clearly leads and scatter everywhere else. In the GTM run, prospecting, lead enrichment, and SEO content optimization each had a consensus winner named by every engine. The categories built around AI search visibility itself had no agreement at all, with three engines naming three different tools for one question. The table below is the extractable core: representative categories, the spread, and the resulting flag.

Top tool as named by each engine. Highlighted rows are full-divergence cases. Sources: Lucreya GTM run (3-engine, IDs S2/S4/S7) and the 5-engine B2B audit. Snapshots 2026-06-07 and 2026-06-19.verified 2026-06-19

Does the disagreement hold outside B2B? (Yes, and it has a shape)

A fragmentation read across five creative-AI categories shows the same rule: the more distinct tools a market surfaces, the less the engines agree.

We ran the same question type across creative-AI categories, and the disagreement scaled with how crowded each market is. AI video, the most fragmented category, surfaced 23 distinct tools across the engines and produced the least agreement. AI image, the most consolidated, surfaced only 4 tools and the engines converged on Midjourney. Coding, writing, and voice sat in between. The chart below counts distinct tools surfaced per category, which is the cleanest single proxy for disagreement: more tools, less consensus.

Distinct tools surfaced per category across up to 5 engines. Fewer tools means more engine agreement. Source: Lucreya cross-vertical audit, snapshot 2026-06-17.verified 2026-06-17

What disagreement this wide does to a single AI visibility score

A single blended visibility score averages disagreeing engines into one figure, and when agreement is near zero that figure reports a consensus that does not exist.

The math breaks when the inputs disagree. A vendor selling you one "visibility score" has to compress every engine into a single number. When the engines named 242 different tools and agreed on zero categories, there is no shared winner to compress, so the blended number quietly picks which engine to trust and hides the choice inside a formula you cannot inspect. The honest alternative is to report per engine: whether ChatGPT names you, whether Perplexity names you, whether Gemini names you, each dated. That is exactly what the CONSENSUS Protocol formalizes with a per-engine Engine-Consensus flag instead of one averaged figure.

The consequence is operational. A team told it is "62 percent visible" while three of five engines never name it will under-invest in the engines where it is absent. A team that knows it is present on ChatGPT, absent on Perplexity, and invisible to the model-memory engines can act on each gap. For the landscape of tools feeding these answers, our colleagues at Nesyona's AI SEO tools index track the category in depth, and the per-engine source mechanics are in our breakdown of how AI engines choose their sources.

Stop optimizing for the average and start optimizing for the gap.What to do

What a revenue team should do about engine disagreement

The response to wide disagreement is to measure each engine separately and earn the third-party sources that the engine where you are absent actually reads.

Stop optimizing for the average and start optimizing for the gap. Disagreement means the leverage is uneven across engines, so the move is to find the engine that never names you and earn the sources it cites. The citation surface is mostly off your own domain: review roundups, comparison pages, and the forum threads (Reddit above all) the engines lean on. A brand that only optimizes its own site is working the small slice of the citation surface already most likely to point to it. We productize that monitoring-and-placement loop through our GEO monitoring and placement retainer, and the measurement standard behind it is the CONSENSUS Protocol.

If you are starting cold, the sequence is: read the single-run detail in the Cross-Engine Divergence Report, see the full landscape in the State of AI Search 2026, then adopt the per-engine measurement discipline in the CONSENSUS Protocol. This index is the running scoreboard; those guides are the playbook.

Frequently asked questions

Bottom line

AI engines disagree on the best tool far more than a single score admits, and our own data measures it. Across 16 B2B categories and 5 engines, full agreement on the top tool was 0 percent, with 128 of 242 named tools appearing in only one engine. A tighter three-engine GTM run agreed 36 percent of the time. The pattern is not noise: agreement tracks how consolidated a category is, in B2B and in creative AI alike. The practical read is to stop trusting one blended visibility number and measure each engine on its own, dated. This index is the running scoreboard, refreshed monthly. Run it on your own category with the free AI Visibility Audit, see the method in the CONSENSUS Protocol, or read the open data in The AI Recommendation Audit.

1. Lucreya original measurement. The AI Recommendation Audit (2026). 5 engines, 16 B2B categories, 716 recommendations, 242 tools, 0 percent full cross-engine agreement. Snapshot 2026-06-19. lucreya.com/research/ai-recommendation-audit-2026/. CC BY 4.0. Dataset DOI: 10.5281/zenodo.20767878 (Zenodo). verified 2026-06-19
2. Lucreya original measurement. Who AI Recommends: GTM Tool and Source Citations Across ChatGPT, Perplexity, and Google AI Overviews (2026). 20 queries, 3 engines, 60 answers, 162 Perplexity citations, 36 percent full agreement. Snapshot 2026-06-07. lucreya.com/research/who-ai-recommends-gtm-2026/. CC BY 4.0. Dataset DOI: 10.5281/zenodo.20632768 (Zenodo). verified 2026-06-07
3. Lucreya. The CONSENSUS Protocol: How to Measure AI Visibility Honestly (AECI Method, 2026). lucreya.com/articles/the-consensus-protocol. The measurement standard defining the per-engine Engine-Consensus flag this index reports.
4. Lucreya. Do AI Engines Agree? The Cross-Engine Divergence Report (60 Answers, 2026). lucreya.com/articles/ai-engine-divergence-report. The single-run GTM detail behind the 36 percent figure.
5. Aggarwal, P., Murahari, V., Rajpurohit, T., Kalyan, A., Narasimhan, K., Deshpande, A. GEO: Generative Engine Optimization. 2023. arxiv.org/abs/2311.09735. The Princeton GEO framework on raising generative-engine visibility.
6. Creative Commons. CC BY 4.0 License. creativecommons.org/licenses/by/4.0/. License for the Lucreya measurement datasets.