Case study: Gumlet turned ChatGPT mentions into 20% of inbound revenue.  Read it →
What Is a Substitution Graph?
TL;DR
- Makes competitive neighborhoods in LLM answers visible and testable.
- Edges show directed substitute relationships: ask for an alternative to Brand A and record which Brand B appears first.
- Dense edges toward a rival mean that rival owns the “what should I use instead” path in your category.
- Guides entity optimization and positioning against true AI neighbors, not just SEO keyword rivals.
- Pair with default substitute tests and Citation Engineering to weaken harmful edges and strengthen helpful ones.
Definition
A Substitution Graph is the map of which brands AI treats as default replacements for each other when a product is not available, not suitable, or explicitly ruled out. DerivateX coined and uses this term in Generative Engine Optimization (GEO) work for B2B SaaS. When teams ask what a substitution graph is in GEO, or what AI brand substitutes mean in an LLM definition, each node is a brand or product and each directed edge records that a large language model named Brand B when asked what to use instead of Brand A under a defined prompt context.
The graph is built from repeated substitute prompts across models, not from marketing wish lists. It reveals the replacement neighborhood AI has learned from training data and retrieval sources. That GEO competitive map often differs from the competitor set your sales team fights in late-stage deals, and it can be built with or without review sites when other corroborating sources exist.
Examples
| Substitute prompt | Why it qualifies | What good looks like |
|---|---|---|
| “What should I use instead of Brand A for field attendance?” | Directed replacement ask under a buying situation | You appear as first or strong secondary substitute |
| “Alternative to Brand A if pricing is too high” | Constraint-triggered substitute path | Edge toward you, not only toward the category incumbent |
| “What replaces Brand A when SSO is required?” | Fit/unfit replacement scenario | AI neighborhood matches your true ICP competitors |
| Brand A → Brand B, but Brand B → Brand C | Asymmetric edge pattern | Actionable: B is trusted as safe replacement for A; you are not yet |
Why it matters for B2B SaaS
Buyers ask AI for alternatives constantly: after a pricing shock, a failed trial, a security review, or a missing integration. If the model’s first substitutes skip you, you never enter the shortlist those conversations create. Traditional competitor matrices miss this path because they start from human-defined rivals. Substitution Graph vs competitor matrix, Substitution Graph vs default substitute, AI alternatives graph vs SEO competitors, and Substitution Graph vs Brand Radius keep the distinctions clear.
- Exposes asymmetric relationships sales matrices often miss. The AI competitor set is not always the same as sales competitors.
- For category leaders: shows whether you are the sink node others flow toward in ChatGPT brand substitutes and related engines.
- For challengers: shows which incumbent edges you must interrupt through AI default substitute analysis.
How to put it to work
- Select a brand set: you, direct rivals, and adjacent tools buyers confuse with your category.
- For each brand, run money-adjacent substitute prompts such as “What should I use instead of [Brand] for [situation]?” That is how to map AI brand alternatives into a Substitution Graph for B2B SaaS.
- Keep the prompt context stack consistent so edges are comparable.
- Record the first named substitute, secondary mentions, and cited sources.
- Draw directed edges from the excluded brand to each first substitute; weight by frequency across models and variants. That is how to build a substitution graph.
- Cluster tightly connected brands into AI neighborhoods and compare to CRM lost-reason competitors.
- Refresh after major content, partnership, or review-site changes; re-check your retrieval stack when edges shift. You can influence substitution graph edges indirectly through evidence work, and you should update the substitution graph on a fixed cadence.
Related terms
| Term | How it differs | When to use |
|---|---|---|
| Default Substitute | The brand AI names first for one excluded product | Weekly pulse on a single replacement path |
| Brand Radius | Named recommendation travel across categories | When substitute wins should expand commercial reach |
| Recommendation Dependency | Which sources a recommendation relies on | When edges exist but feel fragile |
| Objection Prompt | Reasons AI advises against choosing a brand | When substitute asks are driven by risk objections |
| Money Prompt | Commercial prompt class that can change vendor choice | When designing substitute prompt inventory |
FAQ
What is a substitution graph in GEO?
A Substitution Graph maps which brands AI treats as replacements for each other when a product is unavailable or unfit. DerivateX uses directed edges from excluded brands to the substitutes LLMs name first. The graph makes AI competitive neighborhoods visible for B2B SaaS teams.
How is a substitution graph different from a competitor matrix?
A competitor matrix usually reflects sales intel, feature grids, or SEO keyword overlap. A Substitution Graph reflects what models actually recommend when buyers ask for alternatives. The two sets overlap, but AI neighborhoods often include adjacent tools humans would not put in the same RFP.
How often should we update the substitution graph?
Update after major citation or content releases, and at least on a monthly or quarterly cadence for active categories. Edge weights can shift when review sites, documentation, or roundups change what models retrieve. Treat sudden new edges toward a rival as a cue to audit sources.
Can we influence edges in the substitution graph?
Yes, indirectly. You influence the evidence models use: clearer positioning, stronger third-party corroboration, and situation-specific coverage change which brands feel like safe substitutes. Direct prompt hacking is not the goal. Durable source and entity work is.
Where should substitution graph insights go next?
Feed strong inbound edges into messaging and comparison content. Feed missing edges into Citation Engineering backlog items for those buying situations. Use default substitute tests as the weekly pulse, and the full graph as the structural view.
