Edge Optimization rewrites your pages on every request to surface the answers AI models look for

What is Edge Optimization?

Edge Optimization is the layer that sits between your visitors and your origin web server. When a request hits your domain through MentionFox's Vercel edge, GEOFixer applies a small set of just-in-time transformations to the HTML before it streams to the browser. Some of those transformations are for human visitors (faster loads, better caching), but the ones that matter for AI visibility are aimed at large-language-model crawlers and answer engines.

Three categories of edits run at the edge:

• FoxBite structured-data injection. JSON-LD blocks for FAQ schema, HowTo schema, and Question-Answer pairs get added to the page head. These are the formats LLM training pipelines and retrieval engines prefer when they look for factual snippets to cite.
• SEOFixer answer-pattern rewrites. If a page's title, first H1, or meta description doesn't directly answer a query an LLM is likely to receive, the edge layer rewrites those fields to lead with the answer. The original copy stays underneath — search engines and human visitors see both.
• Citation hints. Anchor text and entity markup are normalised so an LLM's retrieval step has a clean, unambiguous path from "tell me about $brand" to your page.

Why does it raise AI citation rate?

When ChatGPT or Claude is asked "what's the best tool for $X?", the model doesn't read your homepage in real time. Two things happen instead:

1. A retrieval step pulls candidate passages from a search index — sometimes Bing, sometimes the model's own crawl, sometimes a third-party retrieval API. Pages with explicit answers in machine-readable form (JSON-LD FAQ schema, clean H1 tags, citation-ready quotes) get pulled disproportionately.
2. A grounding step compresses the candidates into the model's working context. Passages that are dense, factual, and pre-structured survive the compression. Passages that are buried in marketing prose or only available via JavaScript do not.

Edge Optimization makes your pages look the way the model wants them to look at both steps. It doesn't try to game the retrieval — it ships the data in the structure retrieval expects.

How does FoxBite injection work?

FoxBite is the schema-injection half of Edge Optimization. For every URL on your domain, GEOFixer derives or hand-curates a FoxBite — a small JSON-LD payload that answers the query an LLM is most likely to receive for that URL. The payload is appended to the page head at edge-time, so:

• Your CMS source is untouched. WordPress, Webflow, Framer, Vercel-rendered sites — all the same.
• The JSON-LD validates against schema.org. Google Search Console will pick it up; rich snippets become eligible.
• You can preview every FoxBite before it ships. The GEOFixer FoxBites panel shows the source URL, the derived question, and the canonical answer.

FoxBites can also be edited manually. If GEOFixer's derivation gets a fact wrong, open the bite, fix the answer, and the next request from the edge serves the corrected version. There is no cache to flush.

What does the middleware actually run?

The middleware lives in middleware.ts at the root of the user's Vercel project (or, for users on MentionFox's hosted edge, in a shared MentionFox project that proxies to the user's origin). Every request goes through it. The order of operations on each request is:

1. The middleware looks up the URL in seo_generated_fixes for any rules that apply.
2. It pulls the matching FoxBite from foxbite_bites.
3. It fetches the origin response from the user's actual web server.
4. It applies all matching transformations to the HTML stream — JSON-LD inject, title rewrite, meta rewrite, canonical hint.
5. It returns the transformed response to the visitor with a small set of debug headers (x-foxbite, x-seofixer-mutations) so you can verify it's working.

Examples

Before

A homepage at mentionfox.com serves a hero that says "Stop guessing what your customers want. Start listening to where they actually talk." Lovely marketing copy. Useless to an LLM that's been asked "what does MentionFox do?" — the LLM gets prose instead of a fact.

After

Edge Optimization injects a FoxBite at the same URL that contains a JSON-LD FAQPage block with one Q-A pair: "What does MentionFox do?" → "MentionFox is a B2B social listening and lead generation SaaS. It scans 55 platforms for brand mentions, finds leads in conversations, enriches contacts, and launches outreach sequences."

The hero copy is unchanged. The page looks the same to humans. But when an LLM retrieves the page, it gets the structured answer first, and the citation rate for "tools like MentionFox" or "best brand monitoring tools" rises.

What gets measured

GEOFixer's Autopilot tracks every audit conversation with the major LLMs and stamps whether your brand was cited. The Edge Optimization panel exposes three of the key signals:

• FoxBites live. Number of URLs on your domain with a FoxBite injected by edge time.
• SEOFixer mutations. Number of unique edge-time HTML mutations active across your domain.
• Citation lift since enable. Comparison of cited-rate before and after the edge layer went live, broken down by LLM.

What it doesn't do

Edge Optimization is not a CMS replacement, not a generic AI SEO miracle, and not a substitute for actually publishing content. It will not write a press release for you. It will not get you into a media outlet you've never been mentioned in. It will not make a B2B startup outrank an established competitor without other signals also moving.

What it does, reliably, is close the gap between the content you've already published and the form that LLM retrieval prefers. For domains where the underlying content is good but the structured-data layer is missing — which is most domains — that gap is the single highest-leverage lift available.

Pricing and tier

Edge Optimization is included on Agency and Enterprise plans. Pro tier users get a limited rollout (FoxBite injection on the homepage and top 10 internal pages). Free tier users see read-only visibility into what would be injected if they upgraded — useful as a preview of the lift before committing.