> ## Documentation Index
> Fetch the complete documentation index at: https://docs.zued.ai/llms.txt
> Use this file to discover all available pages before exploring further.
# Alignment Score
> How closely your content matches the AI consensus. One of zued's two primary KPIs, alongside the Technical Score.
The **Alignment Score** measures how well the most relevant content on your page covers what AI engines actually tell users about your topics. It's one of zued's two primary KPIs — the core signal for whether your content matches what users are hearing from AI.
A high Alignment Score alone isn't enough: if AI crawlers can't access your pages, the score is irrelevant. That's why zued always pairs alignment analysis with the [Technical Score](/metrics/technical-score).
## How it's calculated
zued compares each AI engine response against the most relevant content chunk on your page and produces a single 0–100 score for how well your content covers what that engine tells users about the topic. URL-level and project-level scores are averages across all prompts.
Alongside the score, zued breaks the comparison into four gap types, so you can see *why* a score is what it is, not just that it's low:
| Gap type | What it surfaces |
| ------------- | ----------------------------------------------------------------------------------------------------------------------------------------- |
| **Coverage** | Sub-topics the AI response covers that your page doesn't address |
| **Structure** | Places where the AI answers in a format (table, steps, direct definition) your page doesn't provide, making it harder to extract and cite |
| **Angle** | Where the AI frames or approaches the topic differently than your page |
| **Evidence** | Claims the AI backs with statistics, sources, or examples that your page states without support |
Coverage carries the most weight because it's the entry condition: an AI engine can't cite content that doesn't address the question. Closing coverage gaps usually moves the score the most.
## Score levels
| Score | Level | What it means |
| :---: | ----------- | ---------------------------------------------------------------------------- |
| > 70 | **Strong** | Your content covers what AI engines tell users about this topic |
| 40–70 | **Partial** | Some gaps — AI engines discuss topics your content doesn't fully address |
| \< 40 | **Poor** | Significant gaps — AI engines present information your content doesn't cover |
## The four gap types in detail
### Coverage
The gap type that moves your score the most. zued identifies the distinct sub-topics in the AI response (e.g. "interest rate", "minimum deposit", "switching process") and checks whether each one is also addressed on your page. Coverage gaps are surfaced individually on each prompt, so you know which sub-topic is missing, not just that "something" is. An engine can't cite a page that doesn't address the question, which is why coverage is the entry condition.
### Structure
Structure asks whether your content can be lifted cleanly into an AI answer. Pages that bury their key points inside long paragraphs surface more structure gaps than pages with clear headings, lists, definitions, and a direct answer in the first sentence under each heading. A format mismatch — the AI answers in a comparison table while your page is prose — shows up here.
### Angle
Angle looks at whether your content approaches the topic the same way the AI does. AI answers often lead with a comparison, a decision shortcut, or a specific use case; if your page covers the same sub-topic from a different starting point, that framing difference is surfaced so you can decide whether to match it.
### Evidence
Evidence flags claims the AI answer backs with statistics, sources, named examples, or worked detail that your page asserts without support. Closing evidence gaps is often what turns a page that's "present but not chosen" into one engines cite, because grounded claims are easier to quote with confidence.
## What influences your score
Several signals make alignment higher or lower beyond raw content quality:
* **Match confidence.** If the prompt loosely matches a chunk on your URL (see [Prompts](/platform/prompts)), the alignment score reflects the comparison against that loose chunk — which is usually weak. A loose match doesn't always mean your content is bad; sometimes it means the prompt belongs on a different URL. The [Topic Health view](/recommendations/overview) helps you spot that pattern.
* **Per-engine variation.** ChatGPT, Gemini, Perplexity, Copilot, AI Mode, and Grok retrieve and summarise content differently. The same page can score Strong against one engine and Partial against another for the same prompt. Per-engine scores are shown alongside the average so you can see where the divergence is.
* **Prompt context level.** Prompts run at three context levels (minimal, moderate, rich). Richer prompts surface deeper sub-topics, so the AI answer covers more ground. A page that scores Strong on minimal-context prompts may show coverage and evidence gaps on rich-context ones — that's a signal to go deeper, not a measurement error.
* **Fact-check accuracy.** Fact-check prompts run on a parallel pipeline (see [fact-checking](/platform/prompts)) and don't feed alignment directly. But when fact-check shows AI engines stating something different from your page, that often shows up as a coverage or evidence gap as well.
## Reading score changes
Alignment is recomputed on every snapshot. When you see a score move week over week, the most common causes:
* **Score went down**: AI engines added a sub-topic that your page doesn't cover (Coverage), engines started citing a competitor whose content is better evidenced (Evidence), the AI started leaning on a figure or example your page doesn't include (Evidence), or the prompt set was changed and now includes prompts your page doesn't fit (loose match).
* **Score went up**: you added the missing sub-topic, you deepened an existing section, you fixed an outdated reference, or AI engines naturally aligned with your existing content as the topic matured.
* **One engine diverges sharply**: that engine has different retrieval behaviour for this query class. Check the per-engine breakdown — sometimes a single engine has gone off-source for a prompt and your alignment isn't actually wrong.
## What you see per prompt
Each prompt in your project shows a breakdown:
* **Coverage gaps** — sub-topics the AI covers that your page doesn't address
* **Format mismatch** — when AI responds in a format your content doesn't support
* **Angle differences** — where the AI frames the topic differently than your page
* **Evidence gaps** — claims the AI supports with data or sources that your page doesn't
* **Quick Win** — the single highest-impact, lowest-effort change for this prompt
A low Alignment Score doesn't always mean low visibility. It means your content has gaps compared to what AI engines tell users. Closing these gaps ensures your content matches what users are actually hearing.
## Per-engine breakdown
Scores are calculated individually per AI engine and then averaged. This means you can have a strong score with Gemini and a poor score with ChatGPT for the same page — each engine retrieves and interprets content differently.
The more engines you run, the richer the per-engine breakdown — each engine retrieves and interprets content differently.
## Topic clusters
When multiple URLs cover the same topic, they form a **topic cluster** (see [Topics & Entities](/platform/topics-and-entities)). Each prompt targets a single URL, but coverage analysis for the whole cluster reveals how your pages work together.
Each URL gets its own coverage score, gaps, and recommendations — even when prompts in the same cluster test similar themes. Your buying guide might score well on information depth but miss structural elements, while a product page might score well on structure but lack the depth that AI engines provide.
Topic clusters also feed into [cross-URL analysis](/recommendations/cross-url-analysis), which looks at how your pages work together as an ensemble.