Citation Optimizer
Ask ChatGPT
Ask Claude
Ask Perplexity
Ask CopilotWhat This Dashboard Does
The Citation Optimizer takes one of your pages and runs it through Spyglasses' model of the ChatGPT citation pipeline, gate by gate. It shows you, as a stoplight, exactly where your content would survive ChatGPT's retrieval and where it would fall out, and it gives you a specific fix for every weakness. When you are ready, it rewrites the page for you and lets you re-score until it is ready to publish.
Where the AI Visibility Rankings dashboard tells you which queries you do and don't show up for, the Citation Optimizer tells you why a specific page does or doesn't get cited, and what to change. For the full model behind the gates, see the Citation Optimizer Methodology.
This release scores against the ChatGPT pipeline. Other platforms are tracked but not yet scoreable; more pipelines arrive in Q3 2026.
One-Time Setup: Crawl and Index Your Pages
Accurate page matching needs each of your citation-relevant pages (homepage, product, and informational pages) crawled and its content indexed. A new property starts with none, so the first time you open the optimizer you'll see a setup banner offering to crawl and index your pages.
Click to start it. The crawl runs in the background and the banner shows progress; an embedding pass after the crawl can take a minute or two, which is expected. This is a one-time step per property. Once it's done, the optimizer can match a query to your closest existing page and detect overlap between your pages (see Overlapping Pages below).
Setup uses no credits. Scoring is free; only rewrites are billed (see Cost).
The Flow: Query → Source → Score → Revise
The optimizer is a stepped wizard. Past runs live in a history list you return to at any time.
Step 1 — Query
Pick the query to optimize for. The picker lists your property's real, tracked fan-outs — the queries ChatGPT's own pipeline generated, captured from your latest AI Visibility report — with the highest-impact ones first and a flag for the ones where your brand doesn't rank yet (your ranking gaps).
You can also enter a free-text query, but it's clearly labeled unverified: it scores exactly the same way, it just hasn't been confirmed as a query the platform actually generates. If your property has no completed AI Visibility report yet, run one first so the optimizer has real fan-outs to work from.
Step 2 — Source
Choose what to score against that query:
- Pick a page — the optimizer ranks your own pages by how well their content matches the query. If your domain already ranks for the query, it defaults to the page ChatGPT would actually retrieve and offers your other close matches as alternates.
- Paste a URL — score any page, including an earned-media placement on a publisher you don't own.
- Paste a draft — score raw markdown content with no live URL, for a page you're still writing.
Pick a page type (product, homepage, informational, press release, or general). It drives the rewrite template later.
Step 3 — Score
The run kicks off and the gate strip lights up as each gate completes: gray (pending) → green (pass) → yellow (warn) → red (fail). Click any gate to open its detail panel with the underlying signals and the specific recommendation(s) for that gate.
A few things to expect:
- A ranking gap is red but not a dead end. If your page isn't in ChatGPT's result pool yet, the SERP gate shows that honestly, and the run keeps going so you still get content guidance plus a view of the competitor pages that do rank.
- Hard gates can stop the run. If a page can't be fetched or is too thin to chunk, the run stops there and the downstream gates show as "not evaluated" rather than failed, so you can see the one blocking reason.
- Pages that already rank get diagnostic framing. If you're already in the top results, the optimizer explains what's working instead of prescribing changes you don't need.
Step 4 — Revise and Export
When you're ready, click Rewrite. The optimizer generates a revised draft grounded in the gate recommendations and your page-type template, along with a revised title, meta description, and page-appropriate structured data (JSON-LD). It never invents facts, statistics, or credentials; it works with what your content already supports.
The revision view shows a before/after comparison and a change log that traces each edit back to the recommendation that motivated it. Copy or export the result as markdown or HTML, then re-score it to measure the lift. A readiness banner tells you whether the content is publish-ready, worth revising again, or has plateaued, so you know when to stop. Re-scoring is unlimited and free.
Overlapping Pages
When the optimizer matches a query to your pages, it also flags overlapping pages — pairs of your own pages so similar that AI search can't tell which one to cite. Overlap is a problem because it splits your authority: instead of one strong page ranking, two weaker pages compete with each other for the same citation, and neither wins cleanly.
When you see an overlap flag, you have two fixes:
- Consolidate the pages into one stronger page, and redirect the weaker one.
- Differentiate their focus so each clearly targets a distinct query.
How it's detected: overlap uses a dual signal. Two pages are flagged only when both their full-content embeddings and their meta-description embeddings are highly similar (each above ~0.92 cosine similarity). The dual check matters: an earlier single-signal version flagged distinct-but-templated pages as overlapping (think a product catalog where every page shares the same chrome). Requiring the meta description to corroborate the body content removed those false positives, so when you see an overlap flag, it's a real one worth acting on.
On heavily templated sites, shared layout can still pull two genuinely different pages closer together than their unique content warrants. Per-section page vectors would make overlap detection even more precise on those sites; that's on the roadmap.
Driving It from an AI Assistant (MCP)
The whole score → rewrite → re-score loop is also available to a connected AI assistant through the Spyglasses MCP server, so you can optimize content in a conversation. The assistant lists your tracked fan-outs, matches a page, scores it, reads the gate results and a readiness verdict, generates a rewrite, and re-scores, all scoped to your own properties and without publishing anything. See Chat with your reports for setup, and the methodology page for the tool list.
Cost
- Scoring is free, including unlimited re-scoring. Score as often as you like.
- A rewrite costs credits. Brand and Agency plans include a monthly allowance of rewrites; beyond that, and on pay-as-you-go, each rewrite is billed per use. Each rewrite includes unlimited re-scoring of that page. See the pricing page for current numbers, and Settings → Billing for your credit history, where rewrites appear as "Citation Optimizer."
Tips
- Start from a gap. The highest-value runs are the tracked fan-outs where you don't rank yet. The optimizer shows you exactly which competitor pages to study and which of your pages is closest to competing.
- Score before you write. Paste a draft in Source to check a page against its target query before you publish, not after.
- Fix overlap first. If two of your pages are flagged as overlapping for the same query, consolidating or differentiating them often does more than any single content edit.
- Let the readiness verdict tell you when to stop. Revising past "publish-ready" or "plateaued" rarely moves the score; the banner is there to save you the credits.
- Re-scoring is free, so iterate. Rewrite, re-score, read the change log, repeat until the gates are green.