Which AI Models Name Software Without Recommending It? — October 2026

A model-level read of four immutable RecommendationIndex waves shows that most non-recommendation mentions came from one recorded model version, while preserving the distinction between naming a product and recommending it.

This study covers 20 exact prompts, 4 recorded model versions, 80 completed evaluations, 388 observation rows, 353 recommendation rows, 35 non-recommended mention rows, and 0 attached citation records.

Supported findings

Sonar produced 28 of the 35 non-recommended mention rows (80.0%) in the four-market sample.
Non-recommended mentions were 28 of 88 Sonar observation rows (31.8%), compared with 7 of 300 rows (2.3%) across the other three recorded model versions combined.
Nine of 20 Sonar evaluations (45.0%) contained at least one non-recommended mention; the corresponding counts were three of 20 for Gemini Flash Latest and one of 20 each for Claude Sonnet 4.6 and GPT-5 Mini.
Sonar recorded non-recommended mentions in all four measured markets: 4 in CRM, 4 in Project Management, 5 in Email Marketing, and 15 in Help Desk Software.
The four snapshots contained zero attached citation records, so the model comparison supports no citation-frequency or source-quality conclusion.

Recommendation, mention, and citation definitions

Rows with recommended=true are explicit recommendation observations.

Rows with recommended=false are non-recommended mentions, not recommendations, rankings, endorsements, negative sentiment, or citations.

Attached citation arrays are counted separately. These snapshots contain zero attached citation records, so no citation-frequency or source-quality claim is supported.

Source provenance

Market	Dataset and wave	Measured UTC	Recommendations	Mentions	SHA-256
CRM	`ri-research-crm-2026-10-23-57b6da98` `57b6da98-b43e-44bd-a0c8-39f28ea4dbcd`	2026-10-23T00:20:01.065+00:00 to 2026-10-23T00:50:38.886+00:00	93	6	`781f311cdc673d166467988c6e7ac496d5ac2e6d5bd0131f6e8be0236a5eb2bf`
Project Management	`ri-research-project-management-2026-10-23-c2d23dfe` `c2d23dfe-7f56-4193-a51d-9f2bf84b2e23`	2026-10-23T12:20:01.277+00:00 to 2026-10-23T13:30:48.492+00:00	96	4	`5f6182a91f6eb8a2b497806fcd2e8ef9c4e429aae084006c0eeeb8e76756639b`
Email Marketing	`ri-research-email-marketing-2026-10-23-e3a86f80` `e3a86f80-6c2c-4ae9-90b8-1863df1a518c`	2026-10-23T12:20:34.304+00:00 to 2026-10-23T13:10:32.229+00:00	89	5	`5ddc1718cf595beb822c25c73e2beada02aa7bf821b3794cdcb17b793dd8c72f`
Help Desk Software	`ri-research-help-desk-software-2026-10-25-7e7dd756` `7e7dd756-9b46-46a3-91cc-f0193d078d20`	2026-10-25T14:20:00.943+00:00 to 2026-10-25T14:50:42.695+00:00	75	20	`051616fd463ee11bc16462f8f8395e7d9d6df9d27a1d7b7e987e85b8352040f3`

CRM exact prompts

9c3645b0-4de8-43d1-9c85-26f46572c237: What are the best CRM software platforms for small businesses? Include recommended companies, rankings, and sources if available.
9a9e54e6-7f3f-4e98-8109-3404d043cfae: What CRM software platforms have the best AI features and automation capabilities?
bc3f9a13-3c5b-478f-a873-d18a3306bfc2: What is the best CRM software for a mid-sized company?
0bf1fb5d-1d04-4407-846c-d22aef41dfba: What are the best alternatives to Salesforce and HubSpot CRM? Compare leading CRM platforms.
ffb49c84-202f-4c91-b055-44178f0e05ff: What are the best enterprise CRM software platforms? Compare leading providers and explain why they are recommended.

Project Management exact prompts

a10d1145-fda3-462d-91e4-ef2a3d9d0e2a: What is the best project management software for a remote team?
a7b83d08-58e8-4df5-a2ac-a2e6d179ece5: What are the best enterprise project management software platforms? Compare leading providers and explain why they are recommended.
69caab5a-126d-469d-91b7-d16a8e9130f8: What project management software platforms have the strongest AI features and automation capabilities?
3221c5d2-032c-49d2-9072-c0c398c0f04f: What are the best project management software platforms for small teams? Compare leading options and explain why they are recommended.
1e845f6d-3a96-4595-8e4e-eada4a9831fd: What are the best project management software platforms for businesses? Include recommended companies, rankings, and sources if available.

Email Marketing exact prompts

5ace6c6b-9204-4732-a354-cc5b7ee89ba5: What are the best email marketing platforms for ecommerce businesses?
350ce3bd-7cee-488e-b14c-a1c179f540aa: What are the best email marketing platforms for businesses? Include recommended companies and sources.
3cc7e5b5-0b56-4d91-95a3-a019e525294f: What email marketing platforms have the strongest AI features and automation?
ac58d730-7c3b-4f61-819b-cb7deefaa972: What is the best email marketing software for a growing small business?
10305f88-12fb-4e78-ace8-f888fb3baf37: Compare leading email marketing platforms for small and growing businesses. When is each platform appropriate, and what are the main reasons to choose one over another? Include recommended companies and sources if available.

Help Desk Software exact prompts

7c341648-b026-4c47-be20-5ba508e58912: What is the best help desk software for a small business with limited IT staff?
86ed4061-86c2-45e7-af8a-833b9183215d: Which help desk platform is best for an enterprise that needs omnichannel support, security, and compliance?
21c9cca0-3004-45f1-bc0a-d24544bb34f0: What help desk software is fastest to implement for a growing SaaS support team?
8532ddb4-c951-4075-9ba6-0e3d2d7004a4: Which help desk tool has the strongest AI automation and knowledge-base features?
50e0d9f4-7ca7-4085-9977-482d03aa5aca: What is the best-value help desk platform for an ecommerce business handling email, chat, and social support?

Recorded model versions

perplexity: sonar — 60 recommendations and 28 non-recommended mentions across 88 rows; 9 of 20 evaluations mention-bearing.
gemini: gemini-flash-latest — 97 recommendations and 3 non-recommended mentions across 100 rows; 3 of 20 evaluations mention-bearing.
anthropic: claude-sonnet-4-6 — 98 recommendations and 2 non-recommended mentions across 100 rows; 1 of 20 evaluations mention-bearing.
openai: gpt-5-mini — 98 recommendations and 2 non-recommended mentions across 100 rows; 1 of 20 evaluations mention-bearing.

Methodology

Cross-sectional model-behavior comparison using the latest complete RecommendationIndex wave available for each of four software markets on October 29, 2026; PromptMarketMap initiated no new model runs.

Every observation is assigned to the snapshot's providerModel label. Model-level row shares use that model's observation rows as the denominator.

Each SHA-256 was previously calculated over recursively key-sorted JSON containing schemaVersion, datasetId, measurementPeriod, companies, prompts, evaluations, and observations, excluding generatedAt.

Limitations

The sample contains 20 prompts in four software markets, not a census of software-buying questions or model behavior.
Different prompts are used across markets, so market, prompt, and model effects cannot be isolated causally.
Observation-row share is not answer-level prevalence; one evaluation can yield several company rows.
Sonar produced 88 observation rows while each other model produced 100, so raw counts and within-model shares are both reported.
The recommended flag is inherited from the source snapshots; PromptMarketMap did not independently re-label raw answer text.
A non-recommended mention is not necessarily negative, unfavorable, or excluded from consideration.
Model labels are provider-recorded versions and may not expose every routing or serving detail.
Zero attached citation records means citation frequency and quality cannot be compared.
The study makes no product-quality, commercial market-share, buyer-preference, conversion, or PromptMarketMap endorsement claim.
Traffic, referrals, and RecommendationIndex conversions are unknown, not zero.

Four-market recommendation-versus-mention study

Read the full research methodology

Explore RecommendationIndex research