A Parallel Web for Bots — Inside Scrunch AI’s Agent Experience Platform
Every query ChatGPT answers with a citation instead of a click chips away at classic SEO. Scrunch AI tackles the problem with simplicity: build a second, invisible version of your site that speaks the native dialect of LLMs — served only to agent crawlers via smart middleware. No redesign, no duplicate-content penalties, just higher “share-of-response” inside generative answers. Below we unpack the tech, compare rivals and map quick wins for Search, Programmatic, Social and E-commerce teams.
Key Facts
| Signal | Detail |
|---|---|
| Dual-Site Architecture | Human site stays intact; Agent Experience Platform (AXP) auto-generates an AI-optimised mirror |
| Middleware Router | Traefik-style proxy fingerprints bot traffic, pipes it to the hidden AI path |
| LLM-Targeted Output | JSON-LD, bullet snippets, schema-rich facts; visual fluff stripped out |
| Founders’ DNA | Ex-Quantcast & Hearsay Systems — deep ad-tech + compliance pedigree |
| Series A | $15 M (Mayfield, Decibel, Bain Capital) to scale monitoring & feedback loops |
Why It Matters for Every Channel
| Channel | Pain Today | Scrunch AXP Fix | Performance Lift |
|---|---|---|---|
| Search | AI Overviews steal clicks | Mirror page surfaces inside answers via structured facts | CTR substitution → brand still cited |
| Programmatic | Cookie loss = weak audiences | AXP logs agent intents → 1P signals enrich bid-models | CPM efficiency ↑ 8–12 % |
| Social | Meta look-alikes starve for fresh traits | Feed high-intent “asked-in-ChatGPT” cohorts back to Meta | CPA ↓, ROAS ↑ |
| E-commerce (AMC) | Static segments & rule-based bids | AXP funnels new high-propensity cohorts to Amazon Marketing Cloud | TACoS ↓, shelf-rank lift |
How Scrunch AXP Works (1-Minute Tech Stack)
flowchart TD UA[LLM Crawler ChatGPT, Gemini] -->|Header + IP detect| Proxy[Scrunch AXP Proxy] Proxy -->|Route /ai/*| AXP[AI-Ready Mirror JSON-LD, FAQ, Tables] Proxy -->|Route /| Human[Normal Website] AXP --> DB[Enterprise KB] AXP --> Monitor[Prompt ↔︎ Response Monitor] Monitor -->|Feedback| AXP
- Auto-Transformer — crawls your public site daily → restructures into headings, lists, HowTo/FAQ schema.
- Smart Rules — redact legal boilerplate, amplify product specs, map sources to knowledge-base.
- Continuous Optimiser — simulates prompts, measures citation share, tweaks structure nightly.
Competitive Bench
| Player | AI-Specific Site | Middleware Routing | Deep Analytics | Who It Suits |
|---|---|---|---|---|
| Scrunch AXP | Yes | Yes | Basic, improving | Brands wanting a set-and-forget tech layer |
| Jellyfish SoM | No | No | Strong dashboards | Insight-first SEO & content teams |
| Profound | No | No | Deep prompt replay | Enterprises needing granular reporting |
| Writesonic GEO | No | No | Content writing + sentiment | SMBs doing quick content rounds |
| Adobe LLM Opt. | No | No | Full DX suite | Fortune 100 on Adobe stacks |
Pros & Cons
| ✔ Pros | ⚠ Cons |
|---|---|
| No redesign — deploy as proxy in days | Still young → limited UI & knobs |
| Keeps human UX untouched | Needs DevOps sign-off for reverse-proxy |
| Elevates citation odds across models | Monitoring depth lags Profound / Jellyfish |
| Adds zero script bloat (good CWV) | Dual-site = two compliance checks |
Strategic To-Dos for Performance Marketers
(DIY playbook – build your own “agent-ready” layer)
| # | Move | What to Do | Why It Works |
|---|---|---|---|
| 1 | Stand-Up an “Agent Mirror” | Spin up a lightweight sub-domain, e.g. ai.yoursite.com, or an edge function that serves a machine-only copy of your top-50 money pages (pricing, specs, returns, FAQs). Keep markup ultra-lean → Markdown → HTML or pure JSON-LD. | LLM crawlers parse fast; no CSS/JS cruft. |
| 2 | Surface Facts, Not Fluff | Pull SKUs, dimensions, policies, ship windows into <table> blocks or Product / FAQ schema. | LLMs quote atomic facts; hero copy gets ignored. |
| 3 | Pipe AI Logs → Media | Log prompts hitting the mirror, tag them research / compare / buy, then push as 1P audiences to DV360, Meta CAPI, AMC. | Bid on the questions shoppers actually ask. |
| 4 | Track “Share-of-Answer” | Measure the % of high-intent queries where your brand is cited by ChatGPT / Gemini / Perplexity / SGE. Use manual panels + Sheets or script vendor APIs. | Citation ≫ blue-link rank in the agent era. |
| 5 | A/B Structured Snippets | Test snippet length (40 vs 80 words), schema depth, heading granularity. Weekly cron: crawl mirror → run test prompts → log which variant wins citations. | Optimise for what the LLM actually lifts. |
Nail these five moves and you gain Scrunch-style leverage without a SaaS bill, conditioning your stack to think agent-first across content, analytics and paid-media feedback loops.
🤖 Quick Demo Prompt
{
"role": "system",
"content": "You are an AI-SEO monitor. \
Reward: +2 if brand cited in top-3 Gemini answers, –1 otherwise. \
Tools: scrunch.scan(url), scrunch.getCitations(domain), scrunch.redact(url,selector), scrunch.promote(url,selector)."
}
{
"role": "user",
"content": "Our ‘Eco-Blend Hoodie’ PDP is missing from Gemini answers about sustainable hoodies under $80. Fix."
}Agent flow
- scrunch.scan → detects missing colour/material schema.
- scrunch.promote the price + fabric bullets into AI version.
- scrunch.getCitations post-recrawl → reward loop confirms citation gained.
Further
- TechCrunch — “Scrunch AI raises $15 M to rebuild the internet for AI consumption” https://techcrunch.com/2025/03/04/scrunch-ai-is-helping-companies-stand-out-in-ai-search/
- Decibel VC — “Why dual-serve architectures win the AI crawler race” https://www.decibel.vc/articles/scrunch-rewriting-the-web-for-ai
- Bain & Co — “Marketing’s New Middleman: AI Agents” https://www.bain.com/insights/marketings-new-middleman-ai-agents/
- Deep Dive: The New AI-Driven Web — Advertising’s Future in an Era of Agents & Attention
Prepared by Performics Labs — translating frontier AI into actionable marketing playbooks.