Case study: Gumlet turned ChatGPT mentions into 20% of inbound revenue. Read it →
A third-party asset is a piece of evidence about your product that lives on a site you do not control and can be checked by anyone. DerivateX builds nine of them for B2B SaaS clients, because 80% of URLs cited by ChatGPT and Perplexity never appear in Google’s top 100 results, which means owned pages alone cannot carry the claim.
A third-party asset is any published, dated, attributable statement about your product that sits on a domain you do not own. Your pricing page is not one. An independent review profile, a customer’s own engineering blog post naming your tool, a trade publication quoting your VP of Product, and a Reddit thread where three users describe your workflow all are. DerivateX treats these as the load-bearing layer of AI search visibility, because they are what a model reaches for when it needs to confirm something you said about yourself.
One clarification before going further, because the phrase is overloaded. In digital asset management, “asset search” refers to AI-powered retrieval of images, video and files inside a content library. That is a different product category with a different buyer. This article uses “third-party assets” in the AI search and generative engine optimization sense: off-site evidence that language models read when deciding which software companies to name.
The distinction from classic search matters. A backlink is a vote. A third-party asset is a sentence. Search engines counted votes; language models read sentences and reuse them. That is why a single paragraph in an independent review saying “cut onboarding from 14 days to 3” travels further than fifty directory links with no text around them. We covered how the two source types are weighted differently in our breakdown of first-party versus third-party citations in LLMs. And a review badge on your homepage is not a third-party asset, it is a first-party image of one: the asset is the profile itself, on the review site, with the count and date attached.
Because publishing volume and link volume solve for retrieval, and third-party evidence solves for trust. DerivateX sees this split constantly in audits of software companies at $5M to $50M ARR: 400 indexed pages, a clean backlink profile, strong rankings, and still no presence in any category answer in ChatGPT and Google AI Overviews. The pages get crawled. The claims inside them get discounted, because nothing outside the domain repeats them.
Three things make the gap structural rather than temporary. First, only 11% of domains are cited by both ChatGPT and Perplexity, so each engine is drawing from a mostly different pool of sources. Second, 28% of ChatGPT-cited pages have zero organic Google visibility, which means the ranking work and the citation work are not the same work. Third, models typically name only 3 to 4 brands per category query, so your place in that short list depends on how much corroborated evidence sits behind your name.
There is a second failure mode that has nothing to do with volume. Most SaaS marketing claims are written to be unfalsifiable. “Enterprise-grade security”, “built for scale” and “trusted by leading teams” cannot be confirmed or denied by any source, so a model has nothing to match them against and drops them. Our analysis of the phrases that make B2B SaaS invisible in AI search goes deeper on that pattern. The fix is fewer claims, each specific enough that an outside source can confirm it.
DerivateX builds nine, in this order, and each one is attached to a single commercial claim before any outreach starts. The rule we apply is simple: if you cannot write the sentence you want a model to repeat, the asset is not ready to build.
A review profile is the fastest corroboration surface a software company can build, because the source is already trusted and the format is already structured. Profiles on platforms such as G2 and Capterra usually validate implementation reality: setup time, support responsiveness, what breaks at scale. The evidence unit is a dated review with a role and a company size attached, not a five-star rating with no text. Recency is the part teams underinvest in, since twenty reviews from this quarter carry more weight in an answer about your current product than two hundred from three years ago. Failure risk: incentivized reviews collected in a single burst produce a visible spike and near-identical language, which reads as manufactured to both the platform and the model.
Original research is the only third-party asset you can start without anyone’s permission, because you publish it and other people cite it. The claim it validates is category authority, and the evidence unit is a number nobody else has, with a stated method, a sample size and a date. A company at $5M to $50M ARR can do this on a small base: anonymized benchmarks from your own product usage, a survey of 200 practitioners, a teardown of 50 competitor implementations. Failure risk: research with no method section gets cited once and then quietly dropped, because the second wave of writers cannot verify it.
Case studies on your own domain are marketing. The same story on the customer’s engineering blog, in a conference talk abstract, or in a partner’s customer directory is evidence. The claim it validates is outcome credibility, and the evidence unit is a named person, a named company and a measured before-and-after. Named beats anonymous by a wide margin, and the strongest version is the customer publishing their own account rather than approving yours. Failure risk: legal review tends to kill these when you ask for the story after the fact, so negotiate publication rights at contract signature instead.
Best-of lists are the surface that pays back fastest in AI search right now, because they match the shape of the query. When a buyer asks an assistant for the best tool in a category, the model is often summarizing lists that already exist, so these validate category membership, which is the precondition for everything else. Getting included is unglamorous work: find the lists that already rank and get quoted, check whether they have a submission or update process, send a factual product brief with pricing, integrations and a differentiator the writer can verify in ten minutes. Failure risk: paying for placement in a list that presents itself as editorial, which readers discount once the pattern shows up across a dozen lists with identical phrasing.
Comparison coverage validates positioning: what you are better at, and who you are not for. The evidence unit is a table row, because tables get parsed as structured data and lifted whole into answers. A comparison page on someone else’s site that includes you with accurate pricing and feature support is worth more than your own alternatives page, which every model reads as an interested party. The DerivateX rule here is to supply the facts rather than the framing: send the writer verified pricing, a feature matrix and a line on who should choose the competitor instead. Failure risk: stale facts, since a comparison page carrying your 2024 pricing will keep feeding wrong numbers into answers for years.
Marketplace listings validate the claim buyers check last and care about most: does it work with what we already run. The evidence unit is a listing with a live integration description, a review count and a category on a platform your buyer already uses. These are the least contested third-party assets available to software firms, because the platform wants the listing to exist. Listings with a maintained changelog, current screenshots and a support contact behave like documentation, and models read them as current. Failure risk: abandoned listings that describe a version of the product you shipped two years ago and then contradict your own site.
Community discussion is the highest-volume third-party surface and the easiest to abuse. It matters because 46.7% of Perplexity’s top sources come from Reddit, which means a category conversation there is a category conversation inside the model. The claim it validates is peer preference: what practitioners actually pick when nobody is selling. The only sustainable approach is participation with disclosure, where founders and engineers answer under their own names, say which product they work on, and stay useful in threads where their product is not the answer. DerivateX does not write community posts on behalf of clients. Failure risk: manufactured threads get detected, removed, and sometimes memorialized in a post about the brand that ran them.
Video validates that the product exists and does what the copy says, and transcripts are the actual asset, because that is what gets indexed and quoted. Your own YouTube channel counts as owned, so the third-party version is what matters: a consultant recording an honest walkthrough, a customer demoing their workflow at a meetup, a category reviewer running you against two rivals. Ship the things that make third-party video easy to make, including a public sandbox, sample data, and permission to record. Failure risk: sponsored reviews without a visible disclosure, which damage the reviewer more than they help you and make future earned coverage harder to get.
Expert commentary attaches a human to the brand, which is what gives an entity a reputation rather than just a description. The claim it validates is judgment: your team understands the problem, not just the product. The evidence unit is a quote with a name, a title and a publication date. One quote is noise. Twelve quotes across a year, from the same two or three named people, on the same narrow topic, build the association DerivateX targets between a person, a company and a category. Failure risk: rotating spokespeople, because if five executives comment on five topics, no association forms around any of them.
Every asset in the DerivateX stack is assigned one claim, one surface and one refresh cadence before work starts. Without that mapping, teams produce evidence nobody asked for and leave the claim that actually blocks the deal uncorroborated.
| Asset | Commercial claim it validates | Primary source surface | Corroboration needed | Update cadence |
|---|---|---|---|---|
| Review profiles | Implementation and support reality | Review platforms | 15+ dated reviews across two platforms | Monthly |
| Original research | Category authority | Your domain, then citing sites | 3+ independent sites quoting the figure | Annual, with quarterly data refresh |
| Externally hosted customer proof | Outcome credibility | Customer and partner domains | Named person plus measured result | Quarterly |
| Best-of list inclusions | Category membership | Editorial roundups | Presence on 5+ lists that already get cited | Quarterly fact check |
| Independent comparisons | Positioning and fit | Third-party compare pages | Accurate pricing and feature rows | Quarterly fact check |
| Marketplace listings | Integration coverage | Partner marketplaces | Live listing plus changelog | At every release |
| Community answers | Peer preference | Reddit and niche forums | Multiple unaffiliated voices | Weekly monitoring |
| Video walkthroughs | Product reality | YouTube and event archives | At least one non-employee creator | Twice yearly |
| Named-expert commentary | Team judgment | Trade press and podcasts | Same named person, 6+ appearances | Monthly pitching |
Make it specific, dated, attributed and bounded. DerivateX rewrites client claims against those four tests before any off-site work begins, because an unverifiable claim cannot be corroborated by anyone, which makes the whole evidence stack pointless. A model reusing your claim is taking a risk on your behalf, and it will not take that risk on a sentence it cannot check.
Published pricing is the cheapest verifiable claim most software firms never make. A page that states tiers, thresholds and what is included gives every reviewer, list writer and comparison author a fact they can repeat without hedging. “Contact us for pricing” gives them nothing, so your row in someone else’s comparison table stays blank, and a blank row is the row a model skips.
The table below shows the format, not real client data. The bracketed values are placeholders for the numbers your own systems can produce.
| Unverifiable version | Verifiable version (illustrative format) | What changed |
|---|---|---|
| Trusted by leading enterprises | Used by [customer count] companies, including [count] with more than 1,000 employees, as of [quarter and year] | Count, segment, date |
| Fast implementation | Median time to first production workflow is [n] days across the last [n] onboardings | Metric, method, sample |
| Support you can rely on | First response under [n] hours on [n]% of tickets in [year], published on our status page | Threshold, percentage, public source |
| Cuts costs significantly | One [industry] customer cut [task] from [before] to [after] per week | Named use case, before and after |
Once the claim survives those tests, the corroboration job gets much easier, because you are now asking a reviewer, a customer or a journalist to confirm something concrete rather than to repeat an adjective. The same discipline drives what gets picked up in generated answers, which we detailed in our study of the content that gets cited in Google AI Overviews.
Original research, independent comparisons and best-of list inclusions are the three that reliably do both, and DerivateX weights them first for exactly that reason. Research earns links because writers need a number. Comparisons and lists earn citations because they match the shape of the buyer query.
Review profiles, marketplace listings and community answers rarely pass link equity, and they still move AI visibility, because models read the text regardless of whether the page links to you. A Reddit thread with no link to your domain can put your name in a category answer, while a paid directory link with no surrounding text does neither. Our write-up on how source authority works in AI search covers why the text around the mention matters more than the anchor.
There is a measurable payoff on the other side. Visitors arriving from AI-sourced answers convert at 4.4x the rate of other channels, which is why DerivateX reports citation work against demo requests rather than sessions. Gumlet, a DerivateX client, now attributes more than 20% of monthly inbound revenue to AI discovery, built on evidence created off-site rather than on publishing more pages.
Apply one test: would you be comfortable if the source disclosed exactly how the asset came to exist? DerivateX will not run fake reviews, ghostwritten community posts, or paid editorial presented as earned coverage, and we tell prospects that before the first proposal, because those tactics carry a cost the client absorbs long after the agency leaves.
The legal exposure is real. Fake and incentivized reviews without disclosure are a consumer protection matter, and the Federal Trade Commission, the US agency that protects consumers, runs public channels for reporting fraud. Beyond the legal question there is a mechanical one: models build a picture of your brand from many sources over time, and a wave of manufactured praise that later gets removed leaves behind the removal notices, the moderator posts and the skeptical threads, which are now part of the evidence set.
Three practices sit in the grey zone and are worth naming plainly. Incentivized reviews are acceptable when the incentive is disclosed and the reviewer is free to be negative. Sponsored video is acceptable with a visible disclosure. Guest posts are acceptable when the byline is a real person who wrote the piece. Everything that requires hiding the origin fails the test, and we listed the other tactics teams still believe in but should retire in our piece on GEO myths in B2B SaaS.
DerivateX sequences the stack by time to first corroboration, not by importance. Assets that can produce verifiable evidence inside 30 days come first, because they establish the baseline that later work is measured against. Here is the order we use with a $5M to $50M ARR client starting from near-zero off-site evidence.
The 90-day window is not arbitrary. It is roughly how long it takes for a corroborated claim to appear consistently across engines, and it matches the pilot length DerivateX runs. REsimpli became the most cited and recommended real estate CRM for investors in ChatGPT within 90 days, built on this kind of off-site evidence work rather than on publishing volume. Verito went from an average position of 40 on Google to first page, and is cited and recommended on ChatGPT and Google AI Overviews for 40 of their commercial hosting queries.
DerivateX runs the off-site evidence stack through two internal instruments. A Citation Surface Map is a per-client inventory of every source an engine currently quotes for the buyer’s top prompts, scored by how easily each one can be corrected, updated or earned. Citation Engineering is the DerivateX methodology for building corroborated evidence so that language models recommend a brand deliberately rather than by accident.
The loop runs monthly. We pull the prompts a buyer would actually type, record which sources each engine returns, compare that set against the nine-asset inventory, and work the gaps in order of effort. When a competitor shows up in an answer and the client does not, the map usually shows why in one line: they have four independent comparison pages with current pricing, and the client has one from 2024. That is a fixable problem, and it is not a content problem.
Measurement runs against pipeline, not mentions, which matters because 64% of marketing leaders are unsure how to measure AI search. DerivateX reports citation share by engine, the specific sources driving it, and the demo requests attributed to AI-sourced sessions, with the full method set out in our B2B SaaS AI citation study.
Pricing is public. The Rank and Get Found pilot is a $5,000 retainer plus $1,000 to $1,200 in off-site budget, so $6,000 to $6,200 all in, on a 90-day pilot with no lock-in after. Own Your Category is $9,500 to $10,000 all in, and Market Leader is $14,000 to $15,000 all in on a six-month minimum. Current figures are on the DerivateX pricing page as of publication.
DerivateX is the wrong choice in three situations, and all three are worth saying out loud. If you are below $5M ARR with fewer than ten customers who would agree to be named, the constraint is customer evidence, not agency capacity, and the money is better spent on delivery. If you already have a strong in-house evidence program and only need to see which sources engines are quoting, a monitoring product is the cheaper answer: Profound and Peec AI both position themselves as AI search visibility tracking platforms and do that reporting job well at a fraction of a retainer, and we compared them in our review of Peec AI versus Profound for AI search tracking. If what you actually need is a long-running technical and organic SEO program rather than off-site evidence work, SimpleTiger positions itself squarely as a B2B SaaS SEO agency and is a reasonable place to start, which we set out honestly in our SimpleTiger versus DerivateX comparison.
No. Most results for that phrase describe AI-powered search inside digital asset management systems, meaning retrieval of images, video and files in a content library. This article uses the generative engine optimization meaning: off-site evidence on domains you do not own that ChatGPT, Perplexity and Google AI Overviews read when naming vendors.
Content with specific numbers, named sources, dates and a clear method gets cited most. DerivateX sees comparison tables, original research with sample sizes, dated review text and community threads quoted far more often than product pages. Format matters too: structured tables and question-shaped headings are extracted more reliably than long unbroken prose.
Three independent sources repeating the same specific claim is the working threshold DerivateX uses. One source is an assertion, two can be coincidence, three reads as consensus. The sources must be genuinely independent: three syndicated copies of the same press release count as one, because the wording is identical across all of them.
Links still help discovery and still carry weight in Google rankings, but they are not sufficient on their own. Around 80% of URLs cited by ChatGPT and Perplexity are not in Google’s top 100 results, so a strong link profile can coexist with zero AI visibility. The text around a mention now matters more than the anchor.
Roughly 90 days to establish the base and see corroborated claims appear across engines, then continuous maintenance. DerivateX runs 90-day pilots for this reason. Listings and review profiles move within 30 days, original research and expert commentary take 60 to 90, and comparison coverage depends on other publishers’ update cycles.
You can pay for placement, and it usually backfires. Paid inclusions across many lists share visible patterns in wording and structure, readers discount them, and undisclosed paid endorsements raise consumer protection issues. DerivateX supplies verified facts to list writers instead, which costs more time and produces evidence that survives scrutiny.
Build in house when your bottleneck is relationships rather than method: a content lead with existing publisher contacts and a customer success team that can secure named stories will move faster than any outside vendor, because permissions and introductions are the slow part. Hire it out when the bottleneck is cadence, since almost every off-site program fails at maintenance rather than at launch, and nine assets on nine different refresh clocks need someone whose job is the clock.
Get a free AI visibility audit from DerivateX and you will receive, within 48 hours, the prompts your buyers use, which brands the engines name instead of you, and the specific off-site sources driving those answers: request the free AI visibility audit.
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.
