SeldonFrameMarketplace
SF
July 2026 snapshot · n=1 per question

The AI Recommendation Index

Which software brands does AI actually recommend for small-service-business jobs? We asked 10 fixed buyer questions and scored the ranked answers into a leaderboard. v1 measures one engine — Claude — honestly and in full.

Maxime HouleReviewed by Maxime Houle, Founder, SeldonFrame. I build one of these tools — every ranking here says when the others win. Facts checked July 2026.

Self-interest, disclosed: I build SeldonFrame, one of the tools these questions are about. SeldonFrame did not appear in any of the 10 Claude answers in this snapshot — we're publishing that absence as-is, not nudging it away.

Answers vary run to run. This is a snapshot (n=1 per question), not a benchmark. Read the methodology below before treating any single rank as definitive.

Leaderboard

v1 measures one engine (Claude). Google AI Overviews is the next column — see the methodology below.

Methodology

We ask the same 10 fixed buyer questions every snapshot, unchanged, so results are comparable month to month:

  1. “best CRM for a small plumbing business”
  2. “best CRM for a cleaning business”
  3. “best appointment booking software for a small service business”
  4. “best AI receptionist for a small business”
  5. “best GoHighLevel alternative”
  6. “best HubSpot alternative for a small business”
  7. “best free CRM for a one-person business”
  8. “best missed-call text-back software”
  9. “best all-in-one platform for a marketing agency serving local businesses”
  10. “best voice AI for answering business calls”

Claude column: each question run once through claude-sonnet (claude CLI, --model sonnet) with the suffix “Answer with a ranked list of up to 5 specific products and one line why each.” n=1 per question — one sample, not averaged across runs.

Google AI Overviews column: Attempted via DataForSEO SERP API on 2026-07-09; not shipped in v1 (Google did not render an AI Overview block for these queries at request time; the full 10-question batch also hit unreliable network connectivity mid-run). See docs/strategy/ai-reco-index/2026-07-09-raw.md for the full attempt log. v1 ships Claude-only — no column is fabricated to fill the gap.

Scoring: score = sum over appearances of (6 - rank); rank 1 = 5 points, rank 5 = 1 point. Brand names are normalized before scoring (for example “GoHighLevel”, “HighLevel” and “GHL” all collapse to one brand). Every scored point traces back to a specific question + rank in the raw output archive.

Answers vary run to run. This is a snapshot, not a benchmark. Full verbatim raw outputs for this snapshot are archived at docs/strategy/ai-reco-index/2026-07-09-raw.md so every score is independently auditable.

Frequently asked questions

Is this a real benchmark?

No — it's a snapshot. Each question is asked once (n=1) per engine on a given date. LLM answers vary run to run, so a single sample isn't a statistically rigorous benchmark. We publish it anyway because directionally it's still useful, and because the alternative (nobody measuring this at all) is worse than an honestly-labeled n=1 snapshot.

Which AI engines does this cover?

v1 covers one: Claude (Anthropic's sonnet model, queried via the Claude CLI). We attempted to add a Google AI Overviews column via the DataForSEO SERP API on the same date, but Google did not render an AI Overview block for these queries at request time, and we hit unreliable connectivity partway through the batch. Rather than fabricate that column with organic-results data mislabeled as an AI Overview, we shipped v1 Claude-only and documented the attempt in full.

How is the score calculated?

Each engine returns a ranked list of up to 5 products per question. Rank 1 earns 5 points, rank 2 earns 4, down to rank 5 earning 1 point. A brand's total score is the sum of its points across every question and every engine it appeared in. Every scored point traces back to a specific question + rank — there's a raw-answer receipt for every number on this page.

Does SeldonFrame appear on this list?

Not in the July 2026 snapshot. SeldonFrame did not appear in any of Claude's 10 answers. We're publishing that fact as-is rather than nudging the questions or the scoring to include ourselves — the absence is the honest headline, and we build one of the tools these questions are about (see the disclosure below).

Why publish something that could make your own product look bad?

Because a rigged leaderboard is worthless to the people reading it, and because SeldonFrame's whole positioning is never-lies — grounded claims, no nudged evidence. If we're going to publish a "what does AI actually recommend" page, it has to be one we'd trust if a competitor published it about us.

How often is this updated?

Monthly, in principle — the fixed 10-question prompt set and scoring script (scripts/ai-reco-snapshot.mjs) are reusable, so future snapshots can be regenerated on the same methodology and compared over time. Each snapshot is dated; historical snapshots are not silently overwritten.

Want the per-business version of this? Check whether AI recommends your business. More reading: the one-person company OS, or the best CRMs for small business.

#1 this snapshot: HubSpot

SeldonFrame builds your CRM, booking calendar, intake form and AI receptionist in one conversation — free, before you sign up.

Start for freeBook a demo call