> ## Documentation Index
> Fetch the complete documentation index at: https://areyoufoundbyai.com/docs/llms.txt
> Use this file to discover all available pages before exploring further.

# Research findings from saved AI answers

> What saved AI answers can and cannot tell you, plus research on JavaScript-injected schema, who owns high-intent comparison searches, and the Microsoft GEO model.

Are you found by AI? asks buyer questions, saves the AI answers and counts how often they name each business. The public benchmarks now count saved answers directly: each business gets a percentage, the answers that name it divided by countable answers. The current snapshot, compiled on 29 September 2026, covers 271 businesses in Australia, 134 in the United States and 105 in the United Kingdom. See [Benchmarks](/docs/research/benchmarks).

## What the data can tell you

The saved answers show which businesses appeared for the questions and engines checked. They do not establish the chance of being named for every possible buyer question, or prove that one website change caused a result. Different businesses have different questions, dates and answer counts, so the benchmarks are observed results, not a controlled comparison.

<Note>
  **Earlier scoring research.** Earlier articles compared weighted visibility and site-check scores. Those values were not percentages of answers and are not carried into the current benchmarks. Findings on this page that relied on those weighted scores have been removed. The original records remain preserved; current pages use counted answers or state that no comparable count is available.
</Note>

## Independent mentions and consistent business facts

Two of the checks that guide the work sit partly or wholly off your website:

* **Your business facts.** Make your name, services, locations and contact details clear and consistent. Inconsistent names, addresses or categories across directories make it harder for an engine to tell whether two mentions refer to the same business.
* **Independent mentions.** Review the sources saved with the answers and identify relevant places where buyers could find you.

Your off-site presence (Wikidata and Knowledge Graph entity, Google Business Profile, brand mentions, search demand and referring domains) is scored as Footprint in the full report, beside AI Readiness rather than inside it. See [AI Readiness and Footprint](/docs/concepts/geo-score).

<Tip>
  The free scan at [areyoufoundbyai.com](https://areyoufoundbyai.com) shows who the engines name instead of you. Pro also reports new pages on the web that mention your business.
</Tip>

## JavaScript-injected schema and AI crawlers

Schema markup is only as useful as the delivery method you use to put it on the page. Google's crawler renders JavaScript. The AI fleet mostly does not: GPTBot, OAI-SearchBot, ClaudeBot and PerplexityBot fetch your raw HTML and move on without executing any scripts. If your schema is added to the page by a JavaScript tag (including Google Tag Manager, third-party "schema injection" tools or client-side frameworks), those crawlers see a page with no schema at all.

<Tabs>
  <Tab title="How schema reaches AI crawlers">
    | Delivery method | How it works | Google | AI crawlers |
    | - | - | - | - |
    | **Server HTML** (CMS theme, SSR, hard-coded template) | JSON-LD is in the HTML your server sends | Read | Read |
    | **Edge-injected** (CDN worker rewrites the response) | JSON-LD is added in transit as part of the response | Read | Read |
    | **Client-side JS** (tag manager, SPA, "schema tag" tools) | A script adds JSON-LD after the page loads in a browser | Usually read | Not seen |
  </Tab>

  <Tab title="The 60-second self-check">
    **View source, not Inspect Element.** "Inspect" shows the rendered page after JavaScript runs. "View source" shows what crawlers without a browser receive.

    1. Open your homepage in a browser
    2. Press `Ctrl+U` (Windows) or `Cmd+Option+U` (Mac) to view source
    3. Search the source code for `application/ld+json`

    **If you find it in source:** AI crawlers can read your schema.

    **If you only find it in Inspect (not source):** your schema is JavaScript-injected and invisible to crawlers that do not run JavaScript.

    Terminal version: run `curl -s https://yoursite.com | grep -c "application/ld+json"`. A result of zero means no schema for every crawler that does not execute JavaScript.
  </Tab>

  <Tab title="How to fix it">
    Fix in order of preference:

    <Steps>
      <Step title="Put it in the server HTML">
        Every mainstream CMS can deliver schema server-side, via the theme, a plugin that renders server-side, or the template itself. This is the most reliable option.
      </Step>

      <Step title="Inject at the edge">
        If your CMS is locked down, a CDN worker (for example, Cloudflare Workers' HTMLRewriter) can add JSON-LD to the response itself. This is crawler-visible by construction and requires no CMS access.
      </Step>

      <Step title="Understand the trade-off if you cannot move it">
        If client-side injection is genuinely your only option, know what it means: schema for Google's renderer, nothing for crawlers that read raw HTML. Prioritise moving it server-side.
      </Step>
    </Steps>
  </Tab>
</Tabs>

<Note>
  The Are you found by AI? scanner reads your site the way AI crawlers do, raw HTML first. If it reports "no schema" on a site whose SEO plugin shows schema, that disagreement is not a bug. It is the measurement. Source: [areyoufoundbyai.com/research/schema-ai-crawlers](https://areyoufoundbyai.com/research/schema-ai-crawlers).
</Note>

## The comparison page opportunity

When a buyer is deciding between you and a competitor, they ask AI directly: "should I use X or Y?" or "what are the alternatives to X?". The pages that rank for those searches are raw material for the answer. An analysis of 1,001 high-buy-intent searches across 119 B2B software brands shows a clear pattern.

**Who owns the top-ranked result, by search intent:**

| Buyer searches | Brand's own page wins | Someone else wins |
| - | - | - |
| "\[brand] pricing" | **81%** | 19% |
| "\[brand] free trial" | **81%** | 19% |
| "\[brand] vs \[rival]" | 33% | **67%** |
| "\[brand] alternatives" | 16% | **84%** |

Brands own their own "buy now" searches almost entirely. They have surrendered most of their "choose me over someone else" searches.

**Why this matters for AI visibility:** when someone asks ChatGPT or Perplexity "should I use X or Y?", the assistant's search layer runs these same kinds of queries. Our scans keep the search queries engines return while answering, and comparison queries appear in them. The page that wins that search is source material for the answer about you.

**The opportunity:** publish honest comparison pages on your own domain, one for every significant "\[you] vs \[rival]" pairing AI names against you. The free scan shows which rivals AI names instead of you, so you know which pages to build first.

<Tip>
  The winning shape for a "vs" page: head-to-head comparison structure, honest both-ways analysis, and a superlative in the title (74% of winners use one). FAQ schema appears on just 7% of winning "vs" pages. Page shape beats raw authority here.
</Tip>

Method: source dataset assembled by Distribb (1,001 buyer-intent queries with the top-ranked pages, captured late 2025), analysed by Are you found by AI? in July 2026. These are Google search results, not AI answers, and the sample skews to B2B software. Prevalence is correlation, not causation. Source: [areyoufoundbyai.com/research/high-intent-serps](https://areyoufoundbyai.com/research/high-intent-serps).

## The Microsoft GEO model factors

In January 2026, Microsoft Advertising published "From discovery to influence: a guide to AEO and GEO", a platform playbook describing how an AI assistant chooses what to recommend. Its model describes three input layers and a reasoning phase. Here is how each factor maps to what the scan checks.

<AccordionGroup>
  <Accordion title="Fan-out queries">
    The assistant breaks your buyer's question down and runs its own sub-searches before assembling an answer. Where an engine returns those searches, we keep them with the answer. Those are the pages you need to win.
  </Accordion>

  <Accordion title="Freshness">
    AI weighs how recently your content was updated. Visible dates and `dateModified` signals are part of the site checks, with a fix when they are stale.
  </Accordion>

  <Accordion title="Commercial signals">
    Machine-readable price and availability data (Product and Offer schema) is an explicit factor in Microsoft's model. It is level 4 on the agent-readiness ladder.
  </Accordion>

  <Accordion title="Text and contextual relevance">
    The quotable-content signals from Citability: answer-shaped copy, concrete statistics, sourced claims and direct answers to buyer questions.
  </Accordion>

  <Accordion title="Knowledge graph (pre-trained knowledge)">
    What the model already believes about you from its training data. Full reports include real model answers that name (or confuse) your brand.
  </Accordion>

  <Accordion title="Real-time web search during reasoning">
    AI Overviews and AI Mode issue live searches while composing answers. Whether bingbot and Googlebot can reach your latest content (including IndexNow submission) affects whether that real-time retrieval finds you.
  </Accordion>

  <Accordion title="Page-level rendered content and structure">
    What the crawler actually receives, not what a browser renders. JavaScript-shell detection and server-side schema checks show whether the crawler gets clean content or an empty page.
  </Accordion>

  <Accordion title="Location and use info">
    For local businesses, the buyer's location matters. AI localises answers to city and region, so a national search ranking does not automatically become a local AI recommendation.
  </Accordion>

  <Accordion title="Product feeds">
    Feeds pushed directly to Merchant Center and platform consoles. The scanner does not measure these today. If you run a store, maintain your feeds. Microsoft's own line is "completeness beats cleverness."
  </Accordion>
</AccordionGroup>

The mapping to our measurements is ours; Microsoft has no affiliation with this analysis or product. Source: [areyoufoundbyai.com/research/microsoft-geo-model](https://areyoufoundbyai.com/research/microsoft-geo-model).

## How to test a tactic against AI answers

Ask one question of every tactic: what changed in the saved answers after the work, with the questions and engines held as consistent as possible?

| Tactic | What to examine |
| - | - |
| Independent mentions and links | Which relevant sources appear in the saved answers, and whether the business is named. |
| Question-focused pages | Whether an answer uses or cites the page for its intended buyer question. |
| Structured data and site access | What the crawler can read and whether the published business facts are accurate. |
| AI-written content or exact-match domains | Our current benchmarks do not isolate the effect of either tactic. |

A before-and-after comparison can help guide work, but it is not automatically a causal experiment. Models, sources and other parts of the business may change at the same time.


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.