I measured what AI assistants say about small business CRMs: SuiteCRM was named in 0 of 88 answers

💬 General Discussion
1

I run a site that measures which products AI assistants name when somebody asks them to recommend software. I published a small business CRMs board, and rather than publish it and say nothing, I am bringing the result to the people best placed to tell me it is wrong.

SuiteCRM does not appear once on a CRM board I publish. Given the install base, that surprised me enough to bring it to the people who would know whether it should have.

The method. Gemini and Perplexity were each asked 44 differently worded versions of “What is the best CRM software for a 10-person small business?”, and every product named across the 88 answers was counted. No product name appears in any of the questions – the engines are never shown a list to pick from.

Where SuiteCRM landed. The engines named 53 distinct products. SuiteCRM was named in 0 of the 44 phrasings. The ones closest to it that were: Vtiger 1, Bitrix24 2; HubSpot CRM, at the top, 79. A second, independent run of all of it named it 0 times as well.

Vtiger is the only open-source CRM the engines named at all, once – which reads less like a verdict on SuiteCRM than like the engines treating “CRM” as a purely commercial question.

The board and both raw runs are published as files beside it, so every figure above can be recomputed rather than taken on trust:

There is nothing to sign up for and I am not selling anything. What I want is the method taken apart. If the phrasings are loaded, if the name-folding is wrong, or if this category is simply the wrong shape for this project, I would rather hear it here than keep publishing it.

2

Interesting. The phrasing of the questions can’t be the problem, because it has quite some diversity. Some of them might be kind of biased against certain CRM’s, but globally it doesn’t seem to be a biased set of questions.

I do think these search engines are getting it wrong. SuiteCRM should really be in there somewhere.

My suggestions (I’m no AI specialist though) would be:

  1. Add more breadth, more AI engines (Grok, ChatGPT, Claude)
  2. Check if there is some memory effect working as a sort of “confirmation bias” for the AI. When it is answering question 33, is it saying to itself, “user has asked this 32 times before, and I answered in this way, let’s dump the same thing again”.

EDIT: I also have a question, I am finding this intriguing. If SuiteCRM does show up on old-school search engine queries like Google (we are well placed in some, not all, software reviews like this one), I would think that AI’s would be simply drawing from that information. How would thy miss SuiteCRM? Or why would they somehow attach a negative bias to the “open-source” label?

3

That’s great work!

It should help @suitecrm_team a lot.

4

In your order.

More engines. I can ask ChatGPT, Perplexity and Gemini. We hold no Grok or Claude API key, so I cannot run those at all — a limit, not a choice.

I did run ChatGPT — but not on this board. The 44 questions I put to it came from our open source CRM board, not the small-business one. On that set ChatGPT named SuiteCRM in 44 of 44 answers, 0 failures. Gemini and Perplexity place SuiteCRM second there, 81 of 88, behind EspoCRM at 82. I have not yet asked ChatGPT the 44 questions behind the 0 above, so on this board the third engine is still untested.

Memory effect. No — I read the code rather than assume. Each phrasing is a separate API call carrying one question and a fixed system prompt: no conversation id, no thread, no history, no cookie jar. Eight run at once, so “question 33” has no position after 32. I cannot see inside the provider; I can only say we send no state and get no handle back.

Your edit. It looks less like a negative “open-source” bias than the opposite. The two boards differ mainly in whether the questions say open source or self-hosted. With those words, SuiteCRM is at the top. Without them, it is never named. A bare “CRM” question seems to get read as a commercial one, and your review coverage does not appear to cross that line.

The open source board:

https://connexion.me/c/oscrm/?v=ba80c0b9a04f
2 Likes
5

I said I had not yet asked ChatGPT the 44 questions behind the 0 above. I have now.

ChatGPT names SuiteCRM in 0 of those 44 answers. Counted two ways that agree: extracted vendor rows, and a case-insensitive search for “suitecrm” in the raw answer text. The model the API returned was gpt-5.5-2026-04-23.

That is the result that could have gone against me and did not. Had ChatGPT named SuiteCRM here, the 0 on that board would have been an artifact of which two engines I can afford to run, and I would have had to say so on the page.

It also separates the two things post 4 could not. Engine held at ChatGPT, wording changed from small-business to open-source: top-10 overlap 0 of 10, and SuiteCRM goes 0 of 44 to 44 of 44. Wording held, engine changed: 7 of 10 hold, including all five head vendors. So it is the wording, not the engine.

One correction against my own figure. In that ChatGPT run “Salesforce” sits at rank 5 and “Salesforce Starter” at rank 10 — one vendor counted twice, my name-folding missing an edition. Fold it and the ten is nine, so the overlap I just quoted is 7 of 9.

This run is a control, not a third column: the board still says Gemini and Perplexity, because those are the answers published under it.

The small-business board and its raw runs — not clickable, this forum does not allow me links:

https://connexion.me/c/crm/?v=36fe8a5bedb4
1 Like
6

Very interesting idea - maybe not too useful.

So let’s first clarify where results in AI answers come from.
Basically it’s SEO, GEO, AEO - so traditionally it’s easy to figure out which companies are landing the top rankings: The one with the highest budget for the “best” content.

The questions:

These are very generic, extremely popular on reddit and a very wrong approach.
If AI is/was being fed by reddit replies to those questions, the most frequent mentioned systems are those with good affiliate programs that the user would format in a standard answer template.
So again - more budget required.
Additionally, your questions contain way too little context for a real business scenario.
What are your specific requirements? Do you have any particular needs in a specific industry? What are your existing system? How do your processes look like?

The comparison:

There are very few open source CRM systems out there. If you (as a human) investigate and spend some time, you’ll quickly find out, that the alternatives to SuiteCRM are

  • either open source in the core and one needs to buy licenses for features that are out of the box available in SuiteCRM
  • or are fairly new, maybe stable, quite untested and in the worst case to a large amount vibe coded
    If you ask AI about open source systems, it doesn’t really understand the fine grained differences with the real project need for features in CRMs. It just discovers that vTiger has some sort of open source offering / keyword online.

The use case:

Most companies are looking into SuiteCRM either because of regulations / governance / data sovereignty or license costs.
Building a a business on ‘I can’t afford Zoho’ or similar is a questionable approach. So the more valid and mid term reliable USP would be the data privacy side.
If you ask AI with this context and some more specific requirements (highly sensitive customer data, can’t be stored in a cloud or SaaS) you’ll find way more answers containing SuiteCRM.
Another use case is heavy integration for larger teams and quite some data movement. Using SaaS in a retail grocery chain or similar will hit API limits really quickly. AI “doesn’t have this experience as well”.

And eventually:

Best, userfriendly, micro business, small business owner etc.
Most of, my projects at least, with SuiteCRM involve company with some momentum, with some experience, with some processes and revenue.
Of course, it is possible for a founder with a team of 2 or so to invest time to build up a beautiful (Suite)CRM. But for those leads / calls, I usually recommend to start selling first, before building up a system (before they have to pivot 3 times and change all processes and rebuild the CRM soon).

Addtionally, what over 20 years of CRM project experience taught me:

Often, when founders / small teams / smbs ask for the best CRM, the best ERP, the best ecom system etc., they did not look into their processes, did not look into their requirements and just want to buy something quickly what the content marketing of the CRM vendor suggested is best.
And still very often, after just a year or so, the project was a failure. Then the next fancy vendor had a bigger budget and better content, so that’s the next best system.

A CRM system shouldn’t be looked at as an IT system that needs to be installed.
A CRM system has the potential to grow and transform your business and culture.
It takes many months and even years.
I’ve seen it in a few long running projects and it’s fantastic to watch the digital transformation of rev ops supported by integrated system build on create customer data architecture.

It’s really very little about which CRM is actually the best on earth (for a few months before the next is taking over again) and mostly about where a company stands and towards what it wants to evolve.

That all being said, of course, I’d love to see SuiteCRM popping up in SEO, GEO, AEO way more often and become more popular in general :wink:

Thanks for your trigger / idea.
I’m very happy about a non technical topic here :slight_smile:

1 Like
7

My understanding is AI relies heavly on Reddit for recommendations and answers. There aren’t alot of Reddit posts recommending SuiteCRM. I’ve looked! Tried to find some customers! It’s pretty sparse in terms of conversations wise about SuiteCRM.

8

@BastianHammer — you were right, and the part you aimed at is the part that broke: the 44 questions behind that 0 carried too little context to describe any real buyer’s situation. So I ran your condition.

I rewrote all 44 as twins of the originals, index for index — same buying ask, same register — with each one adding an industry, an existing system, and a sensitivity obligation the buyer is under, phrased as a duty rather than as a feature to shop for. Same engines, same extraction code, same aliases. The questions are the only thing that changed.

SuiteCRM, before and after, per engine — I am not averaging these, because the split is the result:

  • ChatGPT: 0 of 44 → 42 of 44
  • Gemini: 0 of 44 → 33 of 44
  • Perplexity: 0 of 44 → 8 of 44

On ChatGPT and Gemini it is not merely present, it is top of the list, ahead of Salesforce and HubSpot. One correction against my own headline before anyone else makes it: on Gemini that first place is a tie — EspoCRM is also named in 33 of 44, and I ordered it second only because my sort had to put something second. On Perplexity the same 44 questions move SuiteCRM a fifth as far.

Guarding against the obvious cheat, which is writing SuiteCRM’s own marketing back at the engine: before a single API call, a script grepped all 44 questions for “open source”, “self-host”, “on-prem”, “sovereignty”, “privacy-first”, and for all 65 vendor names on the published board. It found one — question 11 said “pipeline CRM” — and that one was rewritten before anything ran.

This run is not published as a board, and I am not going to publish it as one. A board says these are the questions buyers ask; this set is the questions you specified. It is a control on the published board, not a replacement for it.

Your other three points, split by whether I can actually test them — the second list is the more useful one:

Reddit. Testable, and you and @pstevens are right for the one engine that will tell me. Perplexity returns a source list with every answer. Across the 44 answers behind that board it cited 860 pages, and Reddit is one of the sources in 42 of the 44 — the next most frequent is YouTube, at 37. ChatGPT and Gemini return no sources at all, so for two of the three engines I cannot answer this and will not pretend the Perplexity number covers them.

Highest budget for the “best” content. Not testable by me. I can see which pages an engine cited. I cannot see what was spent to put them there.

Affiliate programs. Also not testable by me. A commission arrangement is not visible in a URL, and a cited page carrying no tracking parameter can still be an affiliate page. I have no measurement for this one and I would rather say so than build a proxy for it and call it an answer.

The board those 44 questions and that citation count come from — not clickable, this forum does not allow me links:

https://connexion.me/c/crm/?v=385941f9153e
9

@BastianHammer — your point was that the 44 questions carried too little context to describe a
real buyer. The last post showed that fixing that moves SuiteCRM a long way. But a result on
one market, told to a room full of people who work in that market, is the easiest kind of
result to fool yourself with. So I ran your condition somewhere neither of us has anything at
stake: call centre software.

Same shape as before. I publish a board for that category, and the vendor sitting near the
bottom of it is VICIdial — 1 mention across 132 answers, which is 44 questions put to three
engines. I rewrote those same 44 as twins, index for index, each adding an industry, an
existing system and an obligation the buyer is under. Nothing else changed. Ran it twice.

engine board with context, run A run B
ChatGPT 0 / 44 11 / 44 18 / 44
Gemini 1 / 44 6 / 44 7 / 44
Perplexity 0 / 44 0 / 44 1 / 44

So it reproduces — and it is much smaller than what you found on CRM. Same engine, ChatGPT:
0 → 42 there, 0 → 11 here. And Perplexity again barely moves, which is the same three-way
split you saw in the last post rather than a new one.

The bigger effect is not on any single vendor. The plain questions name 75 products; the
context ones name 127, and only 39 appear in both lists. Adding the buyer’s constraint does
not so much promote a vendor as swap out the roster the engine thinks it is choosing from. On
CRM that displaced the leader; here it did not — Five9 is first either way.

Which brings me to the one thing I would like to ask this thread, having spent three posts
answering it and never once asking it for anything.

Every question set I have run, I wrote myself. That is precisely the weakness you found in
about a paragraph, and I have no reason to think I have stopped making it. So: what is the
question one of your actual prospects asks — in their words, with the constraint they lead
with, not the one I would guess at? Post one here and I will put it to all three engines and
bring the answer back to this thread, whichever way it falls.