Case study: Gumlet turned ChatGPT mentions into 20% of inbound revenue. Read it →
DerivateX has found that most B2B SaaS brands missing from ChatGPT recommendations do not have a content volume problem. They have an entity, evidence and corroboration problem. With only 3 to 4 brands recommended per AI category query, the shortlist is decided by what independent sources say about you, not by how many blog posts you published last quarter.
DerivateX sees four distinct failure modes, and they need different responses. The first is that the brand does not exist as a resolved entity: the model has heard the name but cannot connect it to a category, a customer type, or a set of capabilities. The second is that the entity resolves but the evidence is thin, so the model knows what you claim and nothing about what anyone else says. The third is that the evidence exists but sits on surfaces the retrieval layer does not reach. The fourth, most frustrating for teams that have done everything right, is that the model has the wrong facts and describes you as something you stopped being two years ago.
Content volume solves none of these. A software company can publish 120 posts in a year, rank for hundreds of terms, and still not appear in ChatGPT when a buyer types “best revenue intelligence tools for a 200-person sales team.” The model is not summarizing your website. It assembles a candidate set from sources it treats as independent, then filters that set against the constraints in the prompt. That is why 28% of ChatGPT-cited pages have zero organic Google visibility. Strong SEO is a real asset and it correlates with citation, but the correlation is loose enough that you cannot treat one as a proxy for the other. If your SEO team has been running a clean program and AI visibility is still zero, that is not a failure of the SEO program, it is a different retrieval system with different inputs.
The context matters too: 40% of Google queries now show Google AI Overviews, CTR drops by 61% when they appear, and 73% of B2B sites lost significant traffic between 2024 and 2025. The distribution has moved, and the levers that move it are mostly not on your own domain.
These four outcomes get collapsed into “AI visibility” and they should not be. DerivateX tracks them separately because each has a different cause and a different fix.
A brand mention is the model saying your name in an answer, in any position, with any framing. Mentions tell you the entity exists in the model’s working set for that topic. A recommendation is the model placing you inside a shortlist in response to a buying prompt, usually with a reason attached. That is the outcome that moves pipeline, because a buyer reading “three options worth evaluating” treats the list as pre-filtered. A citation is the model linking a source URL, which may be your site, a review platform, a community thread, or an analyst post. Citations tell you which surfaces the retrieval layer is actually pulling from. An accurate description is the model getting your category, capabilities and pricing model right when it does mention you, which fails independently of the other three.
| Outcome | What it means | Most common cause when it fails | First lever to pull |
|---|---|---|---|
| Brand mention | Model names you at all | Entity not resolved to a category | Category definition and consistent descriptors |
| Recommendation | Model places you in a shortlist with a reason | No independent evidence supporting the claim | Review ecosystems and third-party comparison surfaces |
| Citation | Model links a source URL | Content not retrievable or not extractable | Docs, structured pages, question-shaped headings |
| Accurate description | Model describes you correctly | Stale or contradictory facts across sources | Consistency pass across every surface that describes you |
A brand can score well on mentions and badly on recommendations. That pattern is common in B2B SaaS categories with an incumbent: the model knows you exist, lists you as an alternative, and reserves the recommendation for the brand with more corroborated proof. Our note on what to do when ChatGPT recommends a competitor instead of you goes deeper on that specific pattern.
DerivateX runs a version of this before any engagement, and you can run the short version yourself in one sitting. The point is to identify which of the four failure modes you have, so you spend the next quarter on the right lever.
Run each prompt three times in separate sessions. Outputs vary, and a single run will mislead you in both directions, so DerivateX treats anything under three consistent runs as noise rather than signal.
The source audit almost always produces the same surprise. The cited URLs are dominated by review platforms, community threads, comparison posts on sites the client does not own, and documentation. Own-domain blog posts appear less often than teams expect.
DerivateX orders these by the ratio of effort to observable movement in recommendation behavior. The first four move the needle fastest for most software firms in the $5M to $50M ARR range. The rest compound more slowly and matter more once the basics are in place.
Models recommend brands they can place. If your homepage says you are a “growth platform” while your G2 profile says “sales enablement” and your LinkedIn page says “revenue operations software,” the entity does not resolve cleanly to any one candidate set. Pick one primary category noun, one buyer type, and one differentiating capability, then write them as a copular sentence and repeat that exact structure everywhere: homepage, about page, review profiles, press boilerplate, founder bios.
This is unglamorous and it is the highest-return change most B2B SaaS teams can make. The edit itself takes about a week of coordination. The observable movement lands 3 to 6 weeks later, once sources refresh, and DerivateX finds that category consistency alone frequently moves a brand from absent to mentioned in broad category prompts, because the model finally has a stable place to file you.
Review platforms are heavily cited in vendor evaluation answers because they are structured, independent and updated. Volume is not the lever people think it is. What matters is whether the review text contains the specific language a buyer uses in a prompt. Fifty reviews saying “great product, great support” are close to useless for a prompt like “which tool handles multi-entity consolidation.” Eight reviews that name that use case will do more.
DerivateX builds the ten-prompt review target list a specific way, and it is not by search volume. The list is drawn from the frozen prompt set, filtered to prompts where a competitor is already recommended and the deciding evidence in the cited sources is review text rather than documentation or a comparison post. Those are the prompts where a review campaign can actually flip an outcome. Prompts lost on missing docs get routed to the docs lever instead. Then the constraint language is extracted from each of the ten and put into the customer request verbatim, rather than asking for a generic rating. Audit your existing profiles on G2, Capterra and anywhere else you are listed for stale category placements and outdated feature lists, because those are being read as current fact.
Third-party corroboration is the mechanism behind recommendation position. A model weighting your own comparison page against an independent one will treat the independent one as stronger evidence. The work is placement, not publication: getting your product into roundups, category guides, and comparison posts written by people who are not you.
Which placements to chase first is where most programs waste budget. DerivateX prioritizes off the Citation Surface Map. The Citation Surface Map is the ranked inventory DerivateX builds of every domain already cited across a client’s frozen prompt set, scored by how often it appears and whether the client is currently named on it. Domains already being read by the engines and currently omitting you come first, ahead of higher-authority publications that never surface in your category’s answers. That inverts the usual outreach list. DerivateX allocates a separate off-site budget to placement for exactly that reason, $1,000 to $1,200 a month on the entry engagement as of 2026 on the DerivateX pricing page, because the work has hard costs a retainer alone hides.
Docs, API references, changelogs and help centers get cited disproportionately in technical evaluation prompts, because they answer specific questions in specific language with no marketing hedging. If your docs sit behind a login, render client-side, or live on a subdomain with no internal linking, you are removing your most extractable content from the retrieval layer.
Make the integration list a crawlable page. Make the API reference public. Write a changelog entry when you ship. When a buyer asks “does [tool] support SAML SSO,” the model needs a page that says yes in those words, and a marketing page promising “enterprise-grade security” will not resolve it.
Pricing is one of the top constraints in vendor shortlist prompts, and “contact sales” removes you from every price-qualified answer. You do not have to publish a full rate card. You do have to publish something numeric: a starting price, a band, a per-seat figure, an implementation range.
DerivateX publishes its own figures for this reason. The entry engagement is $6,000 to $6,200 all in, the mid tier is $9,500 to $10,000 all in, and the top tier is $14,000 to $15,000 all in with a six-month minimum. Whatever a reader thinks of those numbers, they can be retrieved and compared. A brand that hides pricing cannot appear in half the prompts its buyers actually type.
Integration pages do two jobs at once. They create entity associations between your brand and the tools your buyers already use, and they generate corroboration when the partner links back or lists you in their own directory. A mutual listing in a partner’s marketplace is a third-party signal that costs nothing beyond the relationship you already have.
DerivateX sequences these against the compound prompts in the frozen set, not against the partner list sales cares about most. The prompt this wins looks like “what works with Salesforce and handles usage-based billing,” so the integrations that appear inside your buyers’ compound prompts get documented first, even when they are smaller accounts. That prompt filters the candidate set by association before it filters by quality, so brands with thin ecosystem documentation drop out at the first step.
46.7% of Perplexity’s top sources come from Reddit. Community threads are treated as unpolished, high-signal evidence, and they are read at scale. The wrong response is astroturfing, which gets detected, damages the brand, and is not something DerivateX will run. The right response is having your team present as identified people answering questions where your buyers already ask them, plus monitoring existing threads about your category so you can correct factual errors in public.
Engine behavior differs here. Our comparison of how ChatGPT, Claude, Gemini and Perplexity cite B2B SaaS sources covers where each engine leans. Only 11% of domains are cited by both ChatGPT and Perplexity, which means a source strategy tuned to one engine is largely invisible to the other.
Case studies written as narrative do not extract well. Case studies written with a named customer, a named metric, a stated timeframe and a stated starting point do. The model is looking for verifiable claims it can repeat without risk. “Improved efficiency” is not repeatable. “Cut close time from 11 days to 4 days across a 40-person finance team in one quarter” is.
The same rule governs how DerivateX writes about its own work. REsimpli became the most cited and recommended real estate CRM for investors in ChatGPT within 90 days. Gumlet attributes more than 20% of monthly inbound revenue to AI discovery. Those sentences carry a subject, a number and a timeframe, which is what makes them quotable.
The final lever is the one that makes the other eight hold. Contradictory facts across sources are a common reason a model refuses to commit to a recommendation, or describes a brand with a hedge. Founding year, headquarters, funding stage, employee count, product name spelling, category and pricing model should all say the same thing on your site, your review profiles, your Crunchbase entry, your LinkedIn page and your press coverage.
This is also the mechanism behind most incorrect descriptions. If you are being described as a tool you deprecated, there is a stale source somewhere still asserting it, and updating that source does more than publishing a new page that contradicts it. Our guide to fixing brand hallucinations in ChatGPT walks through the source-tracing method.
DerivateX sequences by diagnostic result, not by preference. The sequence below is the one used with software companies in the $5M to $50M ARR band, and it holds for most categories.
| What the diagnostic showed | Fix first | Typical time to observable change |
|---|---|---|
| Model cannot say what you are | Category definition plus consistency pass | 3 to 6 weeks |
| Mentioned but never recommended | Review language plus third-party comparison placement | 8 to 12 weeks |
| Absent from constrained or technical prompts | Docs, integrations and pricing transparency | 4 to 8 weeks |
| Described inaccurately or with stale facts | Source tracing plus category definition | 4 to 10 weeks |
| Present on ChatGPT, absent on Perplexity | Community and independent source presence | 8 to 16 weeks |
These are targets based on what DerivateX has observed across client programs, not guarantees. Recommendation behavior is affected by model updates, index refresh timing and competitor activity, none of which any agency controls. Anyone quoting you a fixed date for a fixed citation outcome is selling something they cannot deliver.
DerivateX measures at the prompt level, tracks the outcomes separately, and reports weekly. 64% of marketing leaders are unsure how to measure AI search, and the reason is usually that they are measuring the wrong unit. Traffic from AI referrals is a lagging, partial signal. Recommendation presence against a fixed prompt set is the leading one.
A word on causation. If your review campaign, your docs rebuild and your comparison placements all ship in the same six weeks and recommendation presence rises, you have correlation across three variables and no clean attribution. DerivateX handles this by staggering interventions where the client’s timeline allows and by tracking which specific URLs enter the cited source set. A partner comparison post that starts getting cited two weeks after it publishes is a much tighter causal chain than a score that moved for unclear reasons.
The scoring layer DerivateX uses internally is the AI Visibility Score. The AI Visibility Score is a composite metric that combines mention rate, average shortlist position and citation share across a fixed prompt set and a fixed engine set. It turns a fuzzy question into a number a marketing leader can take to a board review and defend, and DerivateX reports the underlying prompt-level data alongside it so nobody has to trust a single composite. The mechanics sit inside the DerivateX LLM visibility service.
If you are choosing tooling rather than an agency, several platforms now track this. SE Ranking positions itself as an AI SEO platform with an AI Visibility tracker, and its plan comparison, as listed in 2026, shows five LLMs available for tracking with ChatGPT and Perplexity given as the examples, plus a separate AI Search add-on sold on top of the Core and Growth plans at $89 per month on monthly billing. That is a genuinely reasonable route for a team with in-house capacity that wants monitoring rather than execution, and on pure tracking breadth per dollar it beats paying an agency to report. Our roundup of GEO tools for B2B SaaS compares the options in more depth.
DerivateX will say plainly that content volume is sometimes the bottleneck, and pretending otherwise would be dishonest. There are three situations where it is.
The first is a genuinely new category. If you invented the thing and nobody else writes about it, there is no third-party evidence layer to build on and no comparison surface to appear in. You have to create the vocabulary before anyone can corroborate it, and that means publishing.
The second is a site with fewer than roughly thirty substantive pages. Below that threshold there is not enough surface for a retrieval system to find anything, whatever the quality. Entity work on an eight-page site is optimizing a thing that barely exists.
The third is when your existing content answers no specific question. A blog full of thought leadership with no product documentation, no comparison pages, no pricing detail and no use case pages has a coverage gap, not a corroboration gap. Fill the gap first.
Outside those three cases, the marginal blog post is close to inert for recommendation behavior. A software company publishing eight posts a month with no review strategy, no docs and no independent placements is spending real money on the least effective lever available.
DerivateX is also not the right partner for everyone. If you are below $5M ARR, the $6,000 to $6,200 monthly floor is likely wrong for your stage and a tool plus in-house execution will serve you better. If you want a one-time audit and nothing else, the $3,500 diagnostic delivered in two weeks, priced on the DerivateX pricing page as of 2026, is the honest fit, and it credits in full against month one if you convert within thirty days. If you need a large content production team, an agency built around volume is a better structural match than a founder-led team focused on entity and evidence work.
Stop thinking of ChatGPT as a search engine that reads your website. Think of it as an analyst who has read everything written about your category, trusts independent sources more than vendor claims, and has room for three or four names in the answer. That reframe moves spend from content production toward review campaigns, documentation, placement work and consistency maintenance, and it changes the metric from traffic to shortlist presence, which is the number that precedes a demo request. For the strategy layer above these nine levers, the B2B SaaS AI search visibility guide covers the full program.
Fix entity resolution first, then build third-party evidence. Make your category description identical across your site, review profiles and press. Then run review campaigns in your buyers’ constraint language, publish parseable pricing, open your documentation, and earn placements in comparison content you do not own.
Usually because independent sources support their claim and not yours. With 3 to 4 brands recommended per AI category query, the model favors vendors with corroborated evidence across review platforms, comparison posts and community threads. Check which URLs your own prompt tests cite, then work the surfaces where they appear and you do not.
Category clarity, third-party corroboration, retrievability of specific facts, and consistency across sources. Prompts carrying constraints like company size, integration needs or pricing filter the candidate set before quality is assessed, so brands that hide pricing or lack public documentation drop out early. Volume of owned content is a weak signal.
Weekly, against a frozen prompt set of 40 to 150 buyer prompts, with each prompt run several times per engine. Single runs produce noise because outputs vary between sessions. DerivateX tracks mention rate, shortlist position and cited sources separately, since each one fails for a different reason.
Entity and consistency fixes typically show movement in 3 to 6 weeks. Evidence and corroboration work runs 8 to 16 weeks, because third-party placements and review accumulation take time. These are targets from DerivateX client programs, not guarantees, since model updates and competitor activity affect outcomes nobody controls.
Run the free AI visibility audit and DerivateX will send back, within 48 hours, a prompt-level view of where your brand appears across ChatGPT, Google AI Overviews, Perplexity, Gemini and Claude, plus which of the four failure modes is costing you the shortlist.
VP, SEO & AI Search at DerivateX. We're a B2B SaaS SEO and Generative Engine Optimization agency that engineers AI citations in ChatGPT, Perplexity, Claude, and Gemini and connects them to demo bookings and revenue pipeline.
