The recommendation overlap is tight, and one question returned nothing
ChatGPT, Gemini and Perplexity each named Chargebee, Recurly and Stripe Billing in four of the six subscription billing questions we asked. All six queries produced a readable answer, but the consensus was narrow. The three leaders tied at four mentions, with Maxio at three, and then a steep drop to Paddle, Stax Bill and Vindicia at two apiece. Another 22 products got a single mention across the whole measurement.
What surprised us most was the clean break when we asked for free software. The question “best free subscription billing software” pulled an entirely separate roster: Wave, PayPal Invoicing, Invoice Ninja, Zoho Invoice, Square Point of Sale, Odoo, Abby and Avaza. Not one of the paid recommendations appeared in that answer. The assistants treated “free” as a distinct category, not a pricing tier of the same products. A vendor that straddles both free and paid plans might expect to appear in both lists, but here they didn’t.
Another oddity: the query “subscription billing software recommendations” yielded zero product names from any assistant. The responses drew from reddit.com, zoneandco.com, gartner.com, solidgate.com, g2.com, younium.com and appvizer.com, but apparently none of those pages made an unambiguous product recommendation that the models could latch onto. That blank answer is a signal for any company willing to publish a targeted, highly citable recommendation page.
The sourcing pattern explains much of the result. Reddit.com fed every answer, g2.com and zoneandco.com each fed five of six, and recurly.com appeared in four. These four domains effectively set the menu. G2 and Reddit are intuitive, but zoneandco.com, a site that is not a review platform or discussion forum, fed five answers. It suggests that a single domain with high authority in the assistant’s retrieval index can disproportionately shape the output, even if it isn’t a place buyers usually consult.
Zuora’s presence is instructive. The company appeared as two separate entries, “Zuora” and “Zuora Billing,” each named just once in different questions. So even a well-known name in subscription billing struggled to get repeated mentions. The long tail of single-mention products, including three different Zoho billing tools, Salesforce Revenue Cloud and Younium, shows how fragmented the field is once you step outside the top three.
For Chargebee, Recurly and Stripe Billing, the implication is clear: AI assistants are treating you as the default answer set. For everyone else, the measurement says you need to appear in the Reddit threads and comparison pages that these models crawl. Maxio, at three mentions, can likely close the gap with a targeted content push. The companies named once have a steeper climb, but the absence of any recommendation on the broadest query suggests an opening: a page that explicitly says “use X” in a format the models can extract might change the next set of answers.
No product got six mentions, and the free-software question seeded its own separate list
Chargebee, Recurly and Stripe Billing each appeared in four of the six answers. That tie at four was the ceiling. Maxio landed three mentions, and then the counts drop sharply: Paddle, Stax Bill and Vindicia got two each. No product was named in every answer, which means the assistants never converged on a single default for the category. A customer who asks three different assistants the same question will still hear three different first-picks a chunk of the time.
The shape of the list gets interesting once you move below the top tier. Twenty-two other products were named exactly once across the whole measurement. That cohort includes names a category buyer would recognise immediately: Zuora, Zuora Billing, Salesforce Revenue Cloud, and a handful of Zoho variants (Zoho Subscriptions, Zoho Billing, Zoho Invoice). They also include tiny or niche tools like Abby, Growpay, and Younium. A single mention in a six-question run is almost noise. If one assistant tossed a product into a single answer and the other two ignored it, the result is a product that looks present but carries no recommendation weight.
The “best free subscription billing software” question was the engine behind much of that long tail. It pulled a completely separate roster that never appeared elsewhere: Wave, PayPal Invoicing, Invoice Ninja, Zoho Invoice, Square Point of Sale, Odoo, Abby and Avaza. Each of those products got its single mention in that question and nowhere else. The assistants treated “free” as a filter that invalidates the usual top picks, and they built an entirely new shortlist. That explains why names like Wave or Odoo register at all in a subscription billing measurement, and it also explains why the one-per-question long tail is so large. It is not that the assistants are uncertain about the category. It is that one specific question rewrites the candidate pool.
What is absent from the top clusters matters as much as what appears. Zuora, a company often synonymous with enterprise subscription billing, surfaced only twice if you count Zuora and Zuora Billing as separate mentions, and only once per product. A reader who expected Zuora to sit alongside Chargebee and Recurly at four or five mentions will find that gap jarring. The assistants evidently did not surface Zuora for the generic “best” or “recommendations” questions, except one mention in “what subscription billing software should I use.” Stripe Billing, despite Stripe’s dominance in payments, tied at four but did not pull ahead. No single question produced a response where all three assistants agreed on the same product.
A product named once in a measurement this sparse is a coin flip. A product named in four of the six answers is the closest thing this data set has to a safe recommendation. Chargebee, Recurly and Stripe Billing are the only names a seller can point to and say “the major AI assistants bring me up repeatedly across very different phrasings of the same need.” A seller of one of the 22 single-mention products, or of Zuora’s two separately-named entities, should not assume that the assistants are hostile. Absence from these six questions is not absence from the category. It means that for the phrasing we tested, the assistants either did not fetch that product or fetched it only when a very narrow constraint like “free” was applied. Change the question wording, and the list will shift.
The full list, counted
Every product serper+or:chatgpt,gemini,perplexity named across 6 answers, with how many of those answers named it. The share column is out of the answers we read, not out of the category.
| Product | Named in | Share | Read as |
|---|---|---|---|
| Chargebee | 4 of 6 | 67% | 4 answers |
| Recurly | 4 of 6 | 67% | 4 answers |
| Stripe Billing | 4 of 6 | 67% | 4 answers |
| Maxio | 3 of 6 | 50% | 3 answers |
| Paddle | 2 of 6 | 33% | 2 answers |
| Stax Bill | 2 of 6 | 33% | 2 answers |
| Vindicia | 2 of 6 | 33% | 2 answers |
| Abby | 1 of 6 | 17% | one answer only |
| Avaza | 1 of 6 | 17% | one answer only |
| Growpay | 1 of 6 | 17% | one answer only |
The page mix is a tangle of forum threads, vendor blogs and zero-expertise review hubs
Every single answer the three assistants produced was built at least partly on reddit.com. Reddit posts appeared in all six responses, no matter how the question was phrased. The second most-used source, g2.com, fell just one short at five of six, tied with zoneandco.com, a NetSuite billing partner’s content site that few outside the billing niche would recognise. That pairing tells you a lot about how recommendations get assembled here: the assistants lean on a general forum and then add a stack of vendor-owned pages, with traditional analyst coverage playing a smaller role (gartner.com surfaced in three answers) and pure review sites trailing far behind.
The surprise is how heavily vendor domains like recurly.com, solidgate.com and younium.com fed the answers. Recurly’s own site appeared in four of the six responses, equal to its product’s mention count. A buyer in this category would probably not expect an assistant to pull from a vendor’s marketing page to decide what to recommend, but that is what happened. solidgate.com, a payments orchestration company, and younium.com, another billing vendor, each showed up in three answers, tying with gartner.com. These were not one-off citations from a question about “best subscription billing software” only. solidgate.com fed the answers to “best subscription billing software for small teams,” “what subscription billing software should I use,” and “subscription billing software recommendations.” younium.com fed the former two as well as “most affordable subscription billing software.” The sources are clearly being reused across queries that ask very different things.
Only two of the top eight source domains are what you might call independent review or community sites (reddit.com and g2.com). quora.com appeared just twice, and capterra.com, appvizer.com and upflow.io appeared once each. The rest are vendor sites, product comparison pages run by billing companies, or content marketing pages. For a seller not yet on the list, that pattern is meaningful: presence on Reddit and on a competitor’s blog can matter more than presence on Gartner or G2. For a seller already named, the flip side is that a competitor’s content can end up boosting its own product recommendation indirectly, because the assistant treats the vendor page as an authority source while overlooking the conflict of interest.
One question broke the pattern in a way we cannot fully explain. The query “subscription billing software recommendations” returned no product names at all. Yet it still drew on seven sources: reddit.com, zoneandco.com, gartner.com, solidgate.com, g2.com, younium.com, and appvizer.com. The assistants assembled a list of pages but then produced an answer without picking a single vendor. No other question behaved like that. It is possible the phrasing, which sounds like a request for a general guide rather than a specific tool, made ChatGPT, Gemini and Perplexity hesitate to commit, but the source list is almost identical to the others that did commit. We did not see the same refusal on “best subscription billing software” or “what subscription billing software should I use.”
The free-software question, “best free subscription billing software,” pulled from a completely different set of pages, which helps explain why the product list was so distinct. zapier.com, invoiceninja.com, invoicely.com, zoho.com and capterra.com show up only in that question’s sources. Reddit still fed the answer, but quora.com was the only other domain shared with the paid-software questions. The assistants clearly have a mode where a single word, “free,” triggers a separate corpus of vendor pages and blog roundups about free invoicing or billing tools. It did not mix free and paid sources together, which is why the resulting product lists had zero overlap.
The source list, not the feature list, decides who gets named
All six answers were built at least partly on reddit.com, and five of them drew on g2.com. A product that wants to appear in these AI-generated recommendations does not need to dominate the market. It needs to dominate the pages those two sites surface when someone asks about subscription billing. Chargebee, Recurly and Stripe Billing each hit four mentions because they each show up repeatedly in Reddit threads and G2 grids, and because their own content fills the gaps the assistants go looking for. Recurly’s domain, recurly.com, fed four of the six answers, a number no other vendor-owned domain matched. The correlation there is almost perfectly symmetrical: Recurly got four mentions, recurly.com was a source in four answers. We cannot prove the assistant scooped up the recommendation directly from recurly.com, but the alignment is tight enough that you would be foolish to bet against it.
The second most-used source after Reddit and G2 was zoneandco.com, a content site run by a NetSuite billing partner. It appeared in five answers, yet there is no product called Zone. That means the assistant pulled comparison lists or “top” roundups from that domain and scattered credit across the products those articles name. Maxio, at three mentions, is the most plausible beneficiary. Maxio (born from the merger of SaaSOptics and Chargify) is often paired with NetSuite in billing discussions, and zoneandco.com writes extensively about that exact integration. We do not have the page-level content to confirm that link, but the pattern is hard to ignore: a single partner site fed five answers, and a product that fits its editorial sweet spot got the fourth-highest mention count, ahead of many better-known names.
The free-software question pulled from a completely separate set of domains and produced a completely separate set of products. Zapier.com, invoiceninja.com, invoicely.com, capterra.com and zoho.com each appeared in that answer and nowhere else. The result was a list built from blog posts that round up free tools, not from the Reddit and review-site conversation that shaped every other query. Wave, Invoice Ninja, Zoho Invoice and the rest got exactly one mention, and that mention came from the only question where their name could have surfaced. If an assistant never asks a source that covers free billing tools, those products vanish from every answer in the category.
We can also see where a source showed up repeatedly without converting into mentions for the product it ought to promote. Younium’s domain fed three of the six answers, but Younium was named in just one. That gap suggests the assistant either found the domain through syndicated comparison pages that mention Younium only in passing, or it overrode the source’s preference in favour of something it pulled from Reddit or G2. Recurly’s site, by contrast, seems to have worked harder: four appearances as a source, four mentions. Not all content strategies are equal once they pass through an assistant’s retrieval layer.
The same asymmetry shows up in the long tail. Salesforce Revenue Cloud appeared once, and salesforce.com was a source only in that single answer. Zuora and Zuora Billing each got one mention, but zuora.com does not even appear in the source list for those answers; the assistant likely encountered a Zuora mention inside someone else’s article (younium.com or solidgate.com, say). Zoho’s three entries (Zoho Billing, Zoho Invoice, Zoho Subscriptions) got one mention each, and zoho.com turned up as a source only in the free-software answer, meaning the other two mentions almost certainly came via third-party pages.
What we cannot see from this measurement is just as important. We do not know what the source pages actually said. A product appearing on a page that the assistant cites does not mean the page endorsed it, only that the name was present. We do not know how the assistant weighted its sources. A Reddit comment might have carried more influence than a vendor blog post, or the reverse, depending on the model’s training and the framing of the query. And we cannot say whether a product that got zero mentions would have appeared if the assistant simply fetched a different set of pages from the same domains. The list you see is the intersection of search retrieval, content output and whatever post-processing the assistant applies. Chargebee, Recurly and Stripe Billing landed at four mentions not because an expert panel ranked them, but because their names were repeated often enough across the right places that the assistants had no reason to look further.
Your lead is a Reddit thread, a G2 grid, and someone else’s blog post
If your product is Chargebee, Recurly, or Stripe Billing, you just tied for first place with four mentions in six questions. But look at what you are actually holding. Recurly got named four times, and its own domain showed up in four of the six answers. That means at least some of its recommendations were seeded by content Recurly can control. Chargebee and Stripe Billing also hit four mentions, yet neither vendor domain appears in the source list for this measurement. Their position is built on what other people wrote about them, almost entirely on Reddit threads and G2 category pages. The gap between first place and fifth is not market share or product quality. It is whether a comparison forum thread happened to list you in the first three sentences.
The count of four out of six is not a moat. One question, “subscription billing software recommendations,” returned no product names at all, which means the assistants were unable to pluck a confident answer from the pages they scraped. If that query flips in the next round of indexing, whoever writes the page that fills that gap could suddenly land a mention across all six queries. A single well-placed blog post from Maxio (already at three mentions) or even Zuora (a single mention as Zuora and another as Zuora Billing) could shift the tier entirely. The source diet the assistants are on makes this especially brittle. Reddit.com fed every answer, and g2.com fed five. A product that slips off the first page of a popular Reddit thread or slides a few rows down on G2’s grid can lose its spot without a single change to its feature set.
The free-software question shows how quickly the walls go up. None of the top three from the main list appeared there. Instead, Wave, PayPal Invoicing, and Invoice Ninja grabbed the mentions. So if a competitor adds a free plan and gets named in that separate list, they can start bleeding attention from the paid tier the next time a user asks for “most affordable” or “best for small teams.” There is no progressive build. You get named or you do not, and the difference is a handful of Redditors typing your name in a comment that ranks for exactly the query the assistant is searching.
What would cost you the spot is a shift in those source pages, not a better competitor product. If a rival’s G2 profile gets twenty fresh reviews in the subscription management category, the assistant’s summary might start favouring them tomorrow. If a domain like growpay.com (one mention today) publishes a detailed comparison with a heading that matches the verbatim question “best subscription billing software,” it could start feeding multiple answers the next time the model is invoked. Recurly’s domain advantage is real, but it is also a single site indexed alongside twenty-three other domains. The assistant is indifferent. It just counts. A product that can place content on reddit.com, g2.com, and one vendor-owned site can replicate the pattern that secured four mentions this time. The line between four mentions and one is thinner than the number suggests. It is a few links, a few upvotes, and a single page ranking higher than yours.
If you weren’t named, you are absent from the specific pages the assistants turned to
Not a single assistant named a product that did not already appear somewhere in the measurement’s source list. That is not a rough correlation. It is a hard constraint. If your subscription billing software received zero mentions across all six questions, you are missing from the pages ChatGPT, Gemini and Perplexity actually consulted. The mechanism is not a panel of expert review. It is a small, repetitive set of web pages surfaced by a search API, with the assistants reconstructing what they find there.
The most-used source was reddit.com, which fed every single answer. g2.com and zoneandco.com were each used in five of six answers. Recurly’s own domain showed up in four. Those four alone cover the entire top tier of recommended products: Chargebee, Recurly and Stripe Billing each got four mentions, and Recurly’s own content contributed directly to its position. Chargebee and Stripe Billing reached the same count without any of their own domains appearing as sources, meaning they owed those mentions entirely to Reddit threads, G2 category grids and partner blog posts like those on zoneandco.com. Maxio reached three mentions the same way. So being present on those third-party pages is sufficient, and being absent is disqualifying.
The work of getting named is, in the first instance, an audit of the exact pages that fed these answers. You are not trying to win the whole internet. You are trying to appear in the Reddit discussions that rank when someone searches for “best subscription billing software”, in the G2 grid for the subscription billing category, and on the handful of vendor-comparison posts that Google returns for queries like “subscription billing software recommendations”. For the free-software question the source set shifted: zapier.com, invoiceninja.com, capterra.com and zoho.com all appeared, and the product list changed completely. If you offer a free tier and were not named in that answer, you need to be on those specific pages, not just the paid-software ones.
This does not mean buying placements, chasing backlinks in bulk, or running ads. The assistants used Serper, a search API that queries Google, so the underlying engine is organic search ranking. But your own homepage ranking for “subscription billing software” is not what got Chargebee or Stripe Billing named. They were named because they were mentioned in the content that the search API returned for those queries. That content was overwhelmingly third-party pages: forum threads, review aggregator grids, and comparison posts by billing consultants or NetSuite partners. Only Recurly managed to inject its own domain into the answer stream, and it did so by publishing pages that directly answered the buyer questions the assistants were asking. Those pages then ranked well enough that the search API fed them to the assistants.
If your product is absent, start by querying each verbatim question in the measurement on a clean browser and recording which Reddit threads, G2 pages and comparison posts appear. Open those pages. Search for your product name. If it is not there, that is the bottleneck. The fix is not a generic content marketing push. It is getting your product named in the specific threads and grids the assistants already trust. On Reddit that means authentic community participation. On G2 it means enough verified reviews to appear in the category grid. For comparison posts, it means being included in the articles that rank, which often requires contacting the authors or having a profile solid enough that community-driven lists mention you naturally. None of this is quick, and none of it can be gamed with tactics the assistants cannot parse. The assistants only see what the search API returns, and the search API returns what is already there.
How to run the same measurement for your own category this afternoon
You can replicate this measurement for any software category with nothing more than a browser and a spreadsheet. The method is blunt and entirely manual, and that is precisely why it works. Start by writing six questions. The ones we used were “best subscription billing software,” “best subscription billing software for small teams,” “what subscription billing software should I use,” “best free subscription billing software,” “most affordable subscription billing software,” and “subscription billing software recommendations.” You want a mix of general opinions, a buyer-intent question, a budget-focused variant, and at least one that explicitly asks for free tools. That spread forces every assistant to pull from different corners of its retrieval pipeline.
Next, open three separate browser windows for ChatGPT, Gemini, and Perplexity. Ask each of your six questions in each assistant, running them as fresh queries with no follow-up. That gives you 18 prompt-response pairs. Copy every answer in full, exactly as rendered, into a single document or spreadsheet tab. Do not summarise. The assistants will embed product names inside explanatory paragraphs, sometimes with variant spellings or parent brand names. You will need to read each response line by line and record every distinct product mentioned. Decide ahead of time whether you will merge “Zuora” and “Zuora Billing” into a single entity or treat them separately, and apply that rule across every answer. Document your merge decisions so you do not second-guess yourself on the sixteenth answer.
After products, extract the sources. Look for hyperlinks, in-text mentions like “according to G2,” and footnote-style references at the bottom. Capture only the domain name (reddit.com, not a full thread URL). If the same domain appears twice in one answer, count it once per response. The assistants format their citations differently: Perplexity often lists them separately, Gemini might footnote a handful, and ChatGPT will weave them into the prose. Parsing 18 distinct formats by hand is slow, repetitive work, and by question 14 you will be squinting at domains you have never seen before, checking whether “gocardless.com” actually appeared in the text or you just misread the word “recurly” again.
Now build the tally. Count how many of the six total questions each product appeared in, not how many times it was mentioned within a single answer. A product that gets named three times in one answer but never appears elsewhere still gets credit for only one question. Do the same for source domains. This is the point where manual effort turns into genuine friction. The rhythm is broken because every answer is structured differently, and the judgment calls start piling up. If an assistant mentions a product in a negative comparison, do you still count it? We did, because the buyer query was neutral and the product was still surfaced. You need your own rule and you need to apply it without drifting.
The entire process takes an afternoon, and the bulk of that time is pure clerical double-checking. That tedium is the thing that keeps most people from running it. But the output is a precise snapshot of what a prospect would hear if they asked the same three assistants the same six questions today. No scraping, no API key, no automation: just a browser, a spreadsheet, and the willingness to read 18 documents more closely than the assistants themselves did.