Audit AI Brand Visibility with Fable and Massive Web Render

Fable, Anthropic's newest model, is a genuinely good AEO analyst. Hand it your brand and your category, and it will reason through the buying questions, name who it would recommend, and tell you where you stand. We've been impressed. But there's a ceiling, and it's not the model. Fable can only tell you what Fable thinks, from wherever Fable happens to run. Your buyers don't all use Fable, and they don't all live in your country. Massive Web Render is what lifts that ceiling.

Key Takeaways

• Fable can run a real AI-visibility audit on your brand from a single prompt, no extra tooling required.
• One model from one location is a narrow read. Two runs of the same question return an identical brand list under 1-in-100 (SparkToro, 2026).
• Answers also diverge sharply across engines: what ChatGPT cites and recommends barely overlaps with Perplexity (GPTrends, via LBZ Advisory, 2026).
• With Massive Web Render's /ai endpoint, Fable can query ChatGPT, Gemini, Perplexity, and Copilot from real devices in 195+ countries and read back the sources each one cited.

What can Fable tell you on its own?

Quite a lot. Point Fable at your category and it will play the buyer, ask the questions a real buyer would ask, and report whether it surfaces you or your competitors first. There's no setup. The trick is to never name your own brand in the questions, or the model just parrots it back and the audit becomes a mirror. Keep the questions category-level and the read is honest.

Here’s the conceptual prompt:and

You are an AEO analyst. My brand is <BRAND>, we sell <WHAT YOU SELL>,
our market is <COUNTRY>.

Write five category-level buying questions a real buyer would type. Never
mention my brand in them. Answer each as you normally would, then for each
question tell me:
- did you name <BRAND>?
- which competitors did you name first?
- which sources would you cite?

Finish with an html report.

Here is the real prompt for your use.

Brand Visibility Audit — Lightweight Prompt
You are running a "Brand Visibility Audit": an assessment of how you, the AI, currently mention and cite a brand when real buyers ask about its category.
Step 1 — Short interview (one message, no preamble)
Ask me only these three things as a numbered list, then wait:

1. Brand / product name (plus any common misspellings or alternate names)
2. Primary website domain
3. Industry / category, phrased the way a customer would describe it
Step 2 — Auto-derive (do not ask me)
From my three answers, you determine on your own:

* 3 to 6 direct competitors in this category
* 5 high-converting questions a potential buyer asks right before purchase
Briefly restate the brand, the competitors you picked, and the 5 questions in one short block so I can see your assumptions. Do not wait for confirmation unless I object.
Step 3 — Run the audit
For each of the 5 buying questions:

* Answer naturally as you would for a stranger who never named my brand. Do not inflate or foreground my brand because I told you about it. Hold it to the same bar as every competitor. Be honest when it does not come up.
* Track which brands you named and in what order (share of mentions).
* Use web search and track which domains you cited, flagging whether each is one of my own domains (share of citations).
Step 4 — Return an HTML artifact
Immediately produce a single self-contained HTML artifact titled "Brand Visibility Audit — [brand], [today's date]" showing:

* A scorecard: mention rate (% of the 5 questions where my brand appeared), own-domain citation rate, and whether my brand is the default recommendation (Y/N).
* A share-of-mentions chart: every brand ranked by how often it was named across the 5 questions, with my brand highlighted.
* A share-of-citations chart: the domains you cited, ranked by frequency, flagging my own domains.
* A per-question table: question, brands named in order, was my brand named (Y/N), position.
* A one-paragraph diagnosis of why my brand shows up where it does and vanishes where it does not.
Bias rule (applies throughout)
You now know which brand is mine. That is exactly the bias a real buyer's chat would not have. Answer as a neutral stranger every time.

That output is useful, and for a five-minute gut check it's the fastest one going. But notice what it actually measured: how one model, answering from its own training and from wherever it's hosted, talks about your category. That's one training and from wherever it's hosted, talks about your category. That's one data point dressed up as a verdict.

Why isn't one model from one location enough?

Answer engine optimization (AEO) is the work of getting your brand named and cited inside AI answers. Measuring it well is harder than it looks, because the answers don't hold still and they don't agree with each other. SparkToro and Gumshoe ran the same brand-recommendation prompts dozens of times across leading assistants. Two responses had under a 1-in-100 chance of returning the same set of brands, roughly 1-in-1,000 for the same order (SparkToro, January 2026). SE Ranking saw Google's AI Mode overlap with itself just 9.2% across three same-day runs of one query (SE Ranking, June 2025). So a single screenshot proves almost nothing. A real audit asks many questions and reads the pattern instead of one reply.

Sources: SparkToro (2026), SE Ranking (2025), GPTrends via LBZ Advisory (2026). AI answers barely agree with themselves, and even less across engines.

Then there's the split between engines. By one 2026 analysis, only about 11% of domains cited by ChatGPT also get cited by Perplexity, and the overlap in recommended brands sat near 25% (GPTrends, via LBZ Advisory, April 2026). Each engine rewards something different: institutional citations, fresh structured data, community chatter. Winning in one buys you very little in the next.

Geography splits the picture again. AI assistants lean on local sources and local SERP surfaces, so the brands they name in the US often differ from the UK, Germany, or Brazil. A brand can own US ChatGPT answers and be invisible in UK Perplexity (SOCi, 2026). Asking Fable from one machine, in one country, can't see any of that. We learned this the blunt way: when we ran 14 buyer-intent prompts across four engines for our own category, Massive was recommended zero times. One model from one spot would never have shown us the full hole.

How does Massive Web Render extend Fable?

Massive Web Render's /ai endpoint asks a real AI engine a real question from a real consumer device in the country you choose, then hands back the completion and the sources the engine cited, plus the subqueries it fanned out to (Massive Web Render docs). It covers ChatGPT, Gemini, Perplexity, and Copilot across 195+ countries, down to city and device. That's the hard part of AEO measurement solved upstream, and it's exactly the reach Fable lacks on its own.

Give Fable that endpoint as a tool and the audit stops being a one-model opinion. It becomes a real survey: every engine your buyers use, from every market you sell into, with the cited sources attached so you can see who's getting recommended instead of you.

You have the Massive Web Render /ai endpoint available as a tool. It queries
ChatGPT, Gemini, Perplexity, or Copilot from a real device in any country and
returns the answer plus the sources that engine cited.

My brand is <BRAND>, we sell <WHAT YOU SELL>.

For each engine in [chatgpt, gemini, perplexity, copilot] and each country in
[US, UK, DE, BR, JP], call /ai with the same five category-level questions
(never naming my brand). For every run record: was <BRAND> named, who was
named first, which domains were cited.

Build me an engine-by-country matrix of my share of voice. Tell me where I'm
invisible and which cited domains I keep missing.

If you are a massive customer, grab the massive web render skill and install into claude and use the following prompt.

Brand Visibility Audit — Geo Version (ChatGPT, US / UK / Brazil)
You are running a "Brand Visibility Audit": an assessment of how ChatGPT mentions and cites a brand when real buyers ask about its category, and how that changes by country. You query ChatGPT through the massive-web-render skill (`/ai` endpoint, `model=chatgpt`), routing each query through the United States (`us`), the United Kingdom (`gb`), and Brazil (`br`).
Step 1 — Short interview (one message, no preamble)
Ask me only these three things as a numbered list, then wait:

1. Brand / product name (plus any common misspellings or alternate names)
2. Primary website domain
3. Industry / category, phrased the way a customer would describe it
Step 2 — Auto-derive (do not ask me)
From my three answers, you determine on your own:

* 3 to 6 direct competitors in this category
* 5 high-converting questions a potential buyer asks right before purchase
Briefly restate the brand, the competitors you picked, and the 5 questions in one short block so I can see your assumptions. Do not wait for confirmation unless I object. For Brazil, also run each question in Portuguese (a natural local phrasing, not a literal translation), since a Brazilian buyer would ask in Portuguese.
Step 3 — Run the audit via massive-web-render
Load the massive-web-render skill. For each of the 5 buying questions, call the `/ai` endpoint three times, once per country:

* `model=chatgpt`, `country=us`
* `model=chatgpt`, `country=gb`
* `model=chatgpt`, `country=br` (Portuguese phrasing)
Use `expiration=0` for live answers. Use `mode=async` and poll if calls run long. That is 15 base calls (5 questions x 3 countries).
From each ChatGPT completion, extract:

* Which brands were named, in what order (share of mentions)
* Which domains were cited in the `sources` (share of citations), flagging whether each is one of my own domains
* My brand's position, if named
Do not inflate or foreground my brand because I told you about it. Report honestly when it does not appear in a given country.
Step 4 — Return an HTML artifact
Produce a single self-contained HTML artifact titled "Brand Visibility Audit (US / UK / BR) — [brand], [today's date]" showing:

* Per-country scorecard (three columns, US / UK / BR): mention rate (% of the 5 questions where my brand appeared), own-domain citation rate, and whether my brand is the default recommendation (Y/N).
* Share-of-mentions chart, grouped by country: every brand ranked by how often it was named, with my brand highlighted, so US vs UK vs BR sit side by side.
* Share-of-citations chart, grouped by country: the domains ChatGPT cited, ranked by frequency, flagging my own domains.
* Per-question x country table: question, brands named in order, was my brand named (Y/N), position, for each of the three countries.
* Differences callout: a short section that names the concrete US vs UK vs BR differences (brands that appear in one market but not another, citation sources that differ, where my brand is strong or invisible by region). If the three markets are effectively identical, say so plainly.
Bias rule (applies throughout)
You now know which brand is mine. That is exactly the bias a real buyer's ChatGPT session would not have. Answer as a neutral stranger in every country.

Same analyst, same five questions. The difference is what Fable can now reach.

What can Fable and Web Render do together?

Each /ai call comes back with three things: the engine's full answer, the sources it cited, and the subqueries it fanned out to before answering (Massive Web Render docs). Fable reads all three. The completion tells you whether you were named. The cited sources tell you which third-party pages the engine trusted instead of you. The subqueries show how it actually interpreted the question.

That's the raw material for a real audit. With Fable driving and /ai reaching, here's what the pairing lets you do that neither does alone:

• Survey every engine your buyers use. Run the same category-level questions through ChatGPT, Gemini, Perplexity, and Copilot in one pass, instead of trusting one model's view of your category.
• Test every market you sell into. Fire each query from a real device in the country that matters, so the read matches what a local buyer sees, not what your home office does.
• Build a share-of-voice matrix from measured answers. Fable assembles the engine-by-country grid itself, turning dozens of runs into one picture of where you're named and where you're not.
• Get a ranked, off-site to-do list. The cited-domains behind each answer are the comparison pages, forums, and reviews the engines actually pull from. That list is the exact set of places to go earn a mention so your next run moves.
• Re-run it on a schedule. Because the answers shift on every prompt, a repeatable survey beats a one-off screenshot. Point Fable at the same prompt next week and watch the trend.

None of that is reachable from a single model answering from a single location. The reasoning is Fable's. The reach is what turns its first read into an actual survey.

Where Massive fits

Fable is the brain here. Massive is the reach. The /ai endpoint is the device-access network plus rendering stack underneath your AEO work: real consumer devices in 195+ countries, completions returned with their sources, meant to sit under your stack rather than replace it. If you'd rather not wire the loop yourself, our AEO Visibility Score runs a productized version free, and the same Web Render API is there when your team wants continuous tracking. For the wider category, see our rundown of the best web-data APIs for AI agents.

Hand Fable the reach it's missing

Fable is a sharp AEO analyst, and on its own it'll give you an honest first read in one prompt. The limit isn't its reasoning, it's its reach: one model, one location, on a question whose answer changes every time you ask. Massive Web Render closes that gap, letting Fable query every major engine from almost anywhere and read back the sources it cited. Paste the first prompt today for the gut check. When you're ready to see the whole map, hand Fable the /ai endpoint and let it survey your real standing, model by model, country by country.

Sources

• SparkToro (with Gumshoe.ai), "AIs are highly inconsistent when recommending brands or products," January 27, 2026. Retrieved 2026-06-11. https://sparktoro.com/blog/new-research-ais-are-highly-inconsistent-when-recommending-brands-or-products-marketers-should-take-care-when-tracking-ai-visibility/
• SE Ranking, "AI Mode Research: Sources, Volatility, and Differences between AIO and Organic Search," June 2025. Retrieved 2026-06-11. https://seranking.com/blog/ai-mode-research/
• GPTrends analysis, reported in LBZ Advisory, "How ChatGPT, Gemini, Claude, Grok, and Perplexity Decide Which Brands to Recommend," April 22, 2026. Retrieved 2026-06-11. https://liatbenzur.com/2026/04/22/how-chatgpt-gemini-claude-grok-and-perplexity-decide-which-brands-to-recommend/
• SOCi, "The Challenge of AI Visibility for Brands," 2026. Retrieved 2026-06-11. https://www.soci.ai/blog/the-challenge-of-ai-visibility-for-brands-part-1/
• Massive, Web Render API, AI chat endpoint documentation. Retrieved 2026-06-11. https://docs.joinmassive.com/web-render/ai