Case study: Gumlet turned ChatGPT mentions into 20% of inbound revenue.  Read it →
What Is a Source Deletion Test?
TL;DR
- DerivateX coined the term as a practical probe of Recommendation Dependency and retrieval resilience.
- If the brand vanishes after sources are removed or discounted, the prior answer was citation-thin or overfitted to those URLs.
- If the brand remains with new evidence, the recommendation has broader support in model priors or alternate sources.
- Use on Money Prompts, Default Substitute prompts, and Objection Prompts where stakes are high.
- Feed outcomes into Citation Engineering and third-party corroboration plans.
Definition
A source deletion test is a diagnostic where you strip the cited sources from an AI answer and re-ask the question to reveal what still supports the recommendation. DerivateX coined and uses this term in Generative Engine Optimization (GEO) work for B2B SaaS. When people ask what a source deletion test is, what an AI citation stress check is, or what a GEO citation resilience test means, the test challenges whether a large language model recommendation is anchored to a narrow citation set or backed by wider evidence.
Analysts remove, replace, or explicitly tell the model to ignore the previously cited URLs, then repeat the money prompt under the same Prompt Context Stack. The second answer’s brand list, confidence, and new citations expose residual support. Source deletion vs recommendation dependency, deletion test vs ignoring a site in GSC, and citation stress vs AVS stay distinct.
Examples
| Outcome | What happens | What to do |
|---|---|---|
| Collapse | Your brand or claim disappears after deletion | Corroboration gap — expand independent evidence |
| Substitution | A rival fills the gap | Strengthen your residual stack; study rival layers |
| Persistence | Recommendation remains with new sources or reasoned priors | Maintain and amplify those independent sources |
| Objection path | Harmful claim vanishes when one forum thread is ignored | Prioritize that source path over a generic content blast |
Why it matters for B2B SaaS
Visible citations create false confidence. A tidy list of three links can look authoritative while two are your own pages and one is a thin directory. Deleting or discounting those sources shows whether the model can still justify the pick. Durable AI brand recommendations and Citation Engineering resilience depend on that residual support. A B2B SaaS GEO citation stress test catches brittle wins early.
- Engine literacy: Citation-rich tools make deletion literal; chat-style tools may need “ignore these domains” or browsing-off modes when the model will not ignore sources on command.
- Early fragility: Teams shipping only owned content discover weakness before a competitor outranks those URLs in retrieval. Citations can look strong but collapse. Fix brittle AI recommendations that are owned-only.
- Right fix: Distinguishes thin-source problems from deeper priors that need broader corroboration and clearer entity optimization. Source deletion vs Search Console ignore remains a common confusion to clear.
How to put it to work
- Capture a baseline answer for a fixed money prompt, including citations when shown; record brands and claim language. Prioritize source deletion test money prompts first.
- Run a deletion variant: exclude cited domains, ask without those sources, or remove browsing to those URLs where the product allows. That is how to run a source deletion test and how to strip AI citations and re-ask.
- Keep the prompt context stack identical so you test evidence dependence, not a new situation.
- Compare collapse / substitution / persistence; map persistence rates by cluster into a dependency score so you can measure residual recommendation support and residual recommendation after source deletion.
- Inspect the new Retrieval Stack when fresh citations appear.
- Repeat after major content releases — successful Citation Engineering should raise persistence without requiring the same owned URLs every time. GEO citation resilience for B2B SaaS is the goal.
- Track results beside AI Visibility Score so durability has a metric home. Recommendation dependency source deletion stays the diagnostic pair.
Related terms
| Term | Differs how | When to use |
|---|---|---|
| Recommendation Dependency | The metric; deletion is the probe | Summarizing owned vs third-party reliance |
| Retrieval Stack | Ordered layers before/after deletion | Seeing what replaces removed citations |
| Money Prompt | Commercial question under test | Prioritize deletion tests here first |
| Objection Prompt | Risk narratives to stress | When harmful claims may rest on thin sources |
| Default Substitute | First alternative under exclusion | When substitute wins look citation-thin |
| Search Console URL ignore | Crawl/index controls — different system | Not a substitute for answer-time deletion tests |
FAQ
What is a source deletion test?
A source deletion test removes or discounts the citations behind an AI answer, then re-asks the same question. DerivateX uses it to see what still supports a brand recommendation. Persistence implies broader evidence. Collapse implies brittle dependence on the original sources.
Is a source deletion test the same as ignoring a website in Search Console?
No. Search Console controls affect search crawling and reporting. A source deletion test is an answer-time diagnostic inside AI products. It probes dependence on specific evidence for a recommendation, not indexation status alone.
Which prompts should get source deletion tests first?
Prioritize money prompts, default substitute prompts, and high-severity objection prompts. Those answers change shortlists. Vanity definition prompts matter less unless they feed category narratives competitors exploit.
What if the model will not ignore sources on command?
Adapt the protocol: use a different engine mode, ask for alternatives with explicit domain exclusions, or compare logged-out versus browsing-off settings where available. Document the method so runs stay comparable. Perfect deletion is less important than consistent stress on the cited set.
How do source deletion results change the content roadmap?
Collapse on owned-only citations signals a corroboration gap. Persistence with weak new sources signals a need for higher-quality third parties. Persistence with strong independents means you can maintain and amplify those sources rather than only expanding owned volume.
