Use case

Measure your AI visibility across the questions your buyers ask

Run your buyers' questions on ChatGPT, Claude, Gemini, Perplexity and Google, and count where you are named and cited. Our own baseline, with the data.

Jesse Sibley
By , FounderPublished · Last updated · 5 min read

At a glance

JobA repeatable AI visibility baseline
Toolsaeo_run_ai_answers, get_task, aeo_get_site_readiness
Credits29 a question on six engines; 1,506 for this run
Run52 questions, 6 engines, 311 answers, 26 October 2026
ResultNamed in 3.7% and cited in 10.1% of unbranded answers

The prompt

We ran this on ourselves on 26 October 2026, with our own agent, Claude Code connected to Manifold:

TEXT
Build an AI visibility baseline for Manifold. Here are 52 questions our buyers ask, tagged
by buying stage (discovery, comparison, reviews, purchase) and product area. Ask each one
on ChatGPT, Claude, Gemini, Perplexity, Google AI Overviews and AI Mode. For every answer,
record whether Manifold is named, where, whether manifoldmcp.com is cited, which other
brands are named and which URLs are cited. Estimate the cost first.

The questions ran from how-to asks with no tool in mind ("how can an AI agent do keyword research") through comparisons ("best SEO MCP server") to questions that name us ("is Manifold MCP free").

The calls

StepToolInputCredits
1aeo_get_site_readinessmanifoldmcp.com0
2aeo_run_ai_answers with dry_run10 questions, 6 engines0
3aeo_run_ai_answers52 questions in 6 batches of up to 10, 6 engines, 10 brands to detect1,506
4get_taskeach batch's task id0

The dry run priced each batch of 10 questions at 290 credits before anything was spent. One question on all six engines costs 29 credits; Claude's answers cost the most. One answer of 312 failed at the provider and was not charged.

What came back

311 results, of which 276 held an answer: Google shows no AI Overview or AI Mode answer for some questions. For the 41 questions that do not name Manifold:

Buying stageAnswersManifold namedmanifoldmcp.com cited
Discovery ("how do I…")960%0%
Comparison ("best…", "… alternative")954.2%17.9%
Reviews ("is the DataForSEO MCP server worth it")60%0%
Purchase ("cheapest…", "free MCP server for…")2119.0%23.8%
All questions that do not name us2183.7%10.1%

By engine, on the same questions:

EngineNamedCited
ChatGPT2 of 4113 of 41
Claude5 of 418 of 41
Perplexity1 of 411 of 41
Gemini0 of 410 of 41
Google AI Mode0 of 360 of 36
Google AI Overviews0 of 180 of 18

One of the answers that did name us, from Claude, for "best MCP server for AI visibility tracking":

Manifold has 250 free credits a month, with Pro from $20/mo.

What the agent concluded

  • Cited, not named. ChatGPT cited our comparison articles 13 times but named Manifold as a pick twice. The engines read our roundups and recommend the other tools in them.
  • Nothing for discovery. Not one of 96 answers to how-to questions named or cited us. Those questions are where buyers start, before they know a tool's name.
  • Google's engines never mention us. Gemini, AI Overviews and AI Mode named and cited us zero times on these questions.
  • The name is shared. On the 11 questions that name "Manifold MCP", 12 of 58 answers described a different product, most often a Manifold Markets server.
  • The site is not the blocker. The free readiness check passed every test: AI crawlers are allowed, the content is in the HTML, and the sitemap and the homepage's structured data are in place. Only the pricing page lacked structured data, which we have since added.

Run it on your brand

Write 20 to 50 questions your buyers ask, tag each by stage, and give your agent the prompt above with your brand and up to nine competitors. Ask for the estimate first, run it, and keep the raw results. Run the same questions again in two to four weeks and compare the rates, not single answers. For a smaller first look, the guide to tracking your brand in ChatGPT starts with five questions.

Where to go next

Questions people ask

How many questions should a baseline have?

Enough to cover each buying stage and each product area you care about. Ours has 52, from how-to questions through comparisons to pricing. Fewer questions cost less but swing more from run to run.

What is the difference between being named and being cited?

Named means the brand is in the answer text. Cited means the answer links a page on your site as a source. Engines often cite a page and name other brands from it, as they did with our own roundups.

Why do the answers change between runs?

AI answers are live and not deterministic. Keep the same questions and engines, re-run every two to four weeks, and compare the rates rather than single answers.

Does Manifold keep the answers?

Task results are kept for 30 days. Save the raw results yourself if you want to compare later; our baseline is saved as JSON beside the script that re-runs it.

Jesse Sibley

Jesse Sibley

Founder of Manifold. A growth engineer who has worked with London startups for eight years.

  • 4 min read

    Find SEO quick wins in your own Search Console data

    Ask your agent for the queries where you almost rank. A real run on our own Search Console: the prompt, the calls, the output and what it changed.

  • 4 min read

    Build a B2B lead list with your AI agent

    Ask your agent for companies by what they do, where they are and their size. A real run: the prompt, the call, the list it returned and the rows to check.