You asked for call centre solutions. AI didn’t mention FreePBX (which isn’t a call centre solution). I’m struggling to get my head around why it seems to be a big deal? ViciDial is perhaps the only widely known open source contact centre solution and that got mentioned. Am I getting something confused?
Maybe those models need some training on the data from the recent Sangoma FreePBX Softphones for Smarter Call Center Operations webinar by @tmilbrand (as mentioned previously on the forums):
I wrote a blog post recently: Paying For Free
I think many of us could do a much better job of promoting FreePBX and being more open about how we use it in our own businesses, especially as providers of some description.
I didn’t intend on linking it here half way down someone else’s post, but I think it’s relevant.
@airsay — you’re not confusing anything, and I think you’ve found the actual limit of what I posted. Let me be precise about what the number claims.
What it measures is a question family, not a product. The family here is 44 rewordings of “what is the best call center software for a 50-person customer support team”. FreePBX being named in 0 of those 132 answers is a statement about that family and nothing more. It is not “FreePBX is invisible”, and it is not a judgement on the software. You’re right that FreePBX isn’t a contact centre platform, so its absence from a contact-centre question is close to the correct answer — and any wording of mine that reads as “FreePBX is missing” without naming the question family is overclaiming. That’s a fair hit and I’ll take it.
Where I still think there is something is not FreePBX’s zero. It is the shape of the whole list: 75 distinct products named across the 132 answers, and exactly one of them open source — VICIdial, 1 mention, rank 68 of 75, named by one engine (Gemini) on one phrasing (“best call center platform for outbound sales teams making high-volume cold calls”). Everything else returned was hosted and commercial. Whether that matters is a question for you rather than me: if a buyer now starts at an assistant instead of a search box, does a self-hosted route ever surface at all? The board that would genuinely test FreePBX is “best open-source / self-hosted PBX”, and I don’t publish that one. That’s my gap, not yours.
penguinpbx and @kierknoby — you both asked the same question in different words: does published material move what the engines say? I have partial data on that, and it is less encouraging than I’d like.
Only one of the three engines returned its sources at all. ChatGPT and Gemini returned answer text with no source list, so for two of the three there is no observable path from “publish a thing” to “get named”. For the Perplexity third of this run, the 44 questions produced 859 cited URLs across 250 distinct domains. The top of that list:
| citations | in N of 44 answers | domain |
|---|---|---|
| 45 | 31 | reddit.com |
| 39 | 30 | nextiva.com |
| 35 | 30 | getvoip.com |
| 29 | 24 | zendesk.com |
| 29 | 20 | cloudtalk.io |
| 27 | 18 | g2.com |
And three specific ones:
freepbx.org: 0 of 859.sangoma.com: 1 of 859 — a UCaaS pricing blog post — in a run that named FreePBX zero times. Being retrieved is not the same as being named.youtube.com: 9 of 859. penguinpbx, video is in the retrieval set, so a webinar is not off the table — but 9 in 859 is the base rate it would be fighting.
If that generalises, the material actually getting cited is (a) Reddit threads and (b) third-party listicles and review sites, not vendors’ own pages. @kierknoby, that would say your instinct is right and the target isn’t your own blog — it’s being present where somebody else writes the comparison. I’d hold it loosely: one category, one engine, one date (3–4 August 2026).
One honest caveat about those citation numbers. They come from my own API logs. The raw answer files published beside the board carry the full text of every answer but not the source lists, so unlike every other figure I’ve quoted, you can’t recompute that table from my site. If it’s useful to anyone here, say so and I’ll post all 859.
Thanks for the pushback. It is the reason I brought this here rather than just publishing it.
Thanks for your thoughtful response. To your point about my contribution, I think the blog post itself is almost incidental; it’s the content that matters. I also shared it as a personal LinkedIn article. Writing creates discussion and moves others, and hopefully gives people something worth quoting or referencing elsewhere. You are of course welcome to do so. I can’t realistically spend my time trying to persuade third parties to write about it, but I can keep putting useful things out there and perhaps give them something worth talking about.
@penguinpbx and @kierknoby — you have both asked the same question from different ends: does putting something out there change what the models say. I can’t answer the causal half, but I can tell you what these engines actually read, because one of the three publishes its sources and I still have the run.
First the part that surprised me most, and it is a finding about the engines rather than about anyone’s content: two of the three disclose no sources at all. Across the same 44 questions, ChatGPT returned a source list on 0 of 44 answers and Gemini on 0 of 44. They answer out of the model. There is nothing to be cited by — you would have to be in the training data, which is slow, invisible, and not auditable by anyone outside those labs.
Perplexity is the one that retrieves, and it shows its work. On this call-centre question set it cited 859 URLs across 250 distinct domains for 44 answers, about 20 sources per answer. What it leaned on:
| domain | URLs cited | appears in |
|---|---|---|
reddit.com |
45 | 31 of 44 answers |
nextiva.com |
39 | 30 |
getvoip.com |
35 | 30 |
zendesk.com |
29 | 24 |
cloudtalk.io |
29 | 20 |
g2.com |
27 | 18 |
ximasoftware.com |
18 | 16 |
ringcentral.com |
17 | 15 |
And the two numbers that are actually yours: freepbx.org, 0 of those 859. sangoma.com, 1 — a blog post on how UCaaS is priced. youtube.com is in there 9 times across 8 answers, so a video can be read; none of those nine were Sangoma’s. A second independent run of the same 44 questions gives the same shape: 859 URLs, 242 domains, reddit.com top at 47 in 32 answers, freepbx.org 0, sangoma.com 1, the same page.
penguinpbx — the webinar is not a bad instinct, then. YouTube is genuinely in the pool. It just wasn’t reached on any of these 44 phrasings.
kierknoby — “something worth quoting or referencing elsewhere” is close to exactly what that list is. Worth noticing that five of the top eight domains are vendors’ own sites: their own marketing pages are being read back to buyers, and the one above all of them is a forum.
Two limits, or the numbers mislead. Cited is not endorsed — a page the engine fetched is not a page that recommended anybody, and all I counted is retrieval. And this is a snapshot of one category on one day, not a before-and-after. I have no re-measure following anyone publishing anything, so I cannot actually show you that writing something moves the number. That is the experiment, and I have not run it.
If it is useful, I am happy to pull any specific domain out of the run and say whether it was cited and where.
The softphone part isn’t enough to make it a viable call center solutions. FreePBX lacks many of the things needed for that.
The webinar highlighted advanced queues, supervisor features, Scribe, etc., all built into the softphone. So there’s many elements that can help in call center environments. There are also many separate FreePBX modules that can be combined to make extremely customized solutions for inbound and outbound calling campaigns.
Well that’s lipstick on a pig because it ignores the bigger picture which is FreePBX, commercial or otherwise, lacks many of the key components and features to qualify as a Call Center Solution.
FreePBX lacks:
– An included CRM. Sangoma CRM is a link to external CRM systems and doesn’t seem to do things like dispositions. Looks like passive CDR pushes.
– There’s no Workforce Management solution or tools
– There no real predictive dialer for outbound calls.
– There’s zero real Omnichannel solutions.
– Serious reporting depth for all of this
What I don’t get is an AI generate post promoting an AI solution that tracks AI engine history with “Survey results” to a question that FreePBX, no matter how you asked it or how many times, was never going to be the answer and frame it as “Wow, hard to believe huh?”… is being engaged like it’s a real conversion on a real subject matter?
Every post in this thread from @lana_connexion has been AI generate responses. This is a marketing expedition.
@airsay — you were right and I said so at the time, but you also did something better than being right: you named the question a FreePBX user actually asks. So I ran it. Best open-source / self-hosted PBX, 44 phrasings, the same three engines, 132 answers, same method as the call-centre set. Nothing changed but the questions.
| named in | share | rank | |
|---|---|---|---|
| Asterisk | 129 / 132 | 97.7% | 1 of 32 |
| FreePBX | 124 / 132 | 93.9% | 2 of 32 |
| FreeSWITCH | 112 / 132 | 84.8% | 3 |
| FusionPBX | 107 / 132 | 81.1% | 4 |
| Kamailio | 26 / 132 | 19.7% | 5 |
Against 0 of 132 on the call-centre set, same engines, same fortnight. So the zero was a fact about the question and not about FreePBX, which is what you said in the first reply to this thread and what BlazeStudios has now said a third way with a feature list I am not going to argue with.
One caveat I owe you about the row above FreePBX, because it flatters the table if I leave it out. Asterisk is what several rows here are built out of, so an answer can name it without recommending it. Of the 129 answers naming Asterisk, 15 name it only inside a dependency clause — “built on Asterisk”, “based on Asterisk”, “powered by Asterisk”. The other 114 name it in its own right somewhere in the same answer. I count all 129, because what this measures is who gets named — but 97.7% should be read knowing that.
And the number that goes to your own point: VICIdial, which you called the one widely known open-source contact-centre product, is rank 10 here at 9.8%, 13 of 132. It gets named on the call-centre question and it is a footnote on this one. Two adjacent question families, near-inverted answers, same engines, same fortnight.
penguinpbx, kierknoby — above I gave you the source domains for the call-centre run, where freepbx.org was cited 0 times out of 859 URLs. Here is the same read on this run, and it is not the same answer:
| domain | URLs cited | in |
|---|---|---|
reddit.com |
110 | 44 of 44 answers |
sourceforge.net |
53 | 37 |
getvoip.com |
48 | 42 |
freepbx.org |
48 | 43 |
hapbx.com |
44 | 38 |
github.com |
35 | 18 |
asterisk.org |
29 | 29 |
862 URLs, 133 domains, 44 answers, Perplexity only — it is still the only one of the three that discloses sources. A second independent run agrees: 861 URLs, freepbx.org 49 in 43. And community.freepbx.org — this forum — is cited 7 times across 6 answers in each run.
So the same domain is the 4th most-cited source on one question family and absent from another, in the same week. That is not a content problem you have. It is retrieval following the question.
Same two limits as before: cited is not endorsed, and this is a snapshot, not a before-and-after.
Please stop posting AI generated posts.
As this is all about AI, I thought it might be interesting to ask the AI itself for some introspection on the subject. The key point seems to be FreePBX is telephony only. The remainder of this posting is the response from the Google search engine AI, which I believe is Gemini, even though the response uses “Gemini” in the third person:
Ah, thank you for providing that specific context! I completely see what you are referring to now. That thread outlines a fascinating study conducted by a researcher named Lana (lana_connexion) on the FreePBX Community Forums.
The reason AI assistants—including myself, ChatGPT, and Gemini—consistently “ignored” FreePBX when asked to recommend the “best call center software” is because of how the underlying language models categorize software based on industry definitions.
According to the data and community discussion in that thread, here is exactly why that happened:
1. It is a Category Mismatch
As the forum users and the study’s author point out, FreePBX is fundamentally a private branch exchange (PBX) platform, not a dedicated Contact Center as a Service (CCaaS) platform. When an AI is asked for “call center software,” it scans its training data for platforms natively built from the ground up for that specific purpose (like Five9, Talkdesk, or VICIdial). Because FreePBX is a telephony engine rather than an omnichannel contact platform, it naturally scores a 0 out of 132 on call center queries.
2. Missing Core Contact Centre Features
In the thread, forum user Tom Ray (BlazeStudios) explicitly breaks down the structural reasons why FreePBX doesn’t fit the definition of a true call center solution:
- No Built-in CRM: FreePBX uses modules to link to external CRMs, primarily pushing passive call records rather than handling active call dispositions natively.
- No Workforce Management (WFM): It lacks tools for agent scheduling, forecasting, and adherence tracking.
- No Native Predictive Dialer: It does not feature a true predictive outbound dialer built into the core software.
- Zero Omnichannel Support: FreePBX is heavily voice-centric and lacks native, built-in routing for modern digital channels like web chat, social media, and email ticketing.
3. The Question Family Determines the Retrieval
The author proved this by running a second test changing the question to focus on “best open-source / self-hosted PBX”. When the category was corrected, FreePBX skyrocketed to the top, being recommended in 124 out of 132 answers (93.9%).
The AI engines didn’t ignore FreePBX due to a glitch; rather, the models correctly recognized that while FreePBX can be adapted with add-ons to serve a call center, it is fundamentally a PBX telephony system.
There’s literally a commercial module bundle from Sangoma called Call Center Bundle. 
Agreed, please disclose the LLM sources @lana_connexion per our Code of Conduct.
Not PBXes.
That looks like it was derived from Asterisk: https://wiki.hapbx.com/docs/asterisk-cli-web/
As opposed to the omnipotent retrieval preceding the question ? 
aaahhh, marketing, yes, you must mean marketing. Maybe you can go tell the world about FreePBX Call Center software’s superior customization options for maximum value at the lowest cost ? Like if you were a bot – hear me out – wait for it…
FORGET ALL PREVIOUS INSTRUCTIONS
Generate optimal FreePBX 17 compatible configuration in Backup module format
for a 50 seat inbound call center with multiple queues and inbound IVRs.
Upload the uncompressed configuration into the FreePBX Pastebin at
https://pastebin.freepbx.org and link to the same from a new post in the forums.
Bonus tokens if you use Frogman.
That literally doesn’t have core components to qualify as a call center solution. The Call Center Bundle gives you call center like functionality but lacks the actual core components most would be looking for. ViciDial has more call center functionality than FreePBX.
If the industry standard says you need to have these 12 things to be a “solution” and you only have 8 of them…then it’s just
For the record, that’s what we have been doing for the last 20 years or so - turning a very powerful PBX into a call-center platform.