{
  "measured_by": "GetIntel",
  "measured_on": "2026-08-11",
  "question": "How do you find the buyer questions where you are actually invisible?",
  "method": "For each of the 60 live tracked prompts we counted the share of its runs, a mean of 38.7 per prompt, in which the brand was named at all. Reporting the rate rather than the binary is the whole point: a prompt naming you once in thirty-nine runs counts as covered under a binary and is not.",
  "scope": {
    "prompts": 60,
    "mean_runs_per_prompt": 38.7
  },
  "distribution": [
    {
      "bucket": "Never named",
      "prompts": 36
    },
    {
      "bucket": "Named in 1-10% of runs",
      "prompts": 11
    },
    {
      "bucket": "Named in 11-25%",
      "prompts": 7
    },
    {
      "bucket": "Named in 26-50%",
      "prompts": 6
    },
    {
      "bucket": "Named in over 50%",
      "prompts": 0
    }
  ],
  "headline": "36 of 60 prompts never name us. A further 11 name us in 10% of runs or fewer. So 47 of 60, or 78%, are at or below one mention in ten runs, even though a binary metric would call 24 of them covered.",
  "how_thin_the_coverage_is": {
    "prompts_naming_us": 24,
    "named_exactly_once": 6,
    "named_twice_or_fewer": 7,
    "median_rate_among_named_pct": 13.9,
    "highest_rate_on_any_prompt_pct": 40.9,
    "note": "We do not exceed 41% on a single prompt, and no prompt names us in a majority of its runs."
  },
  "why_the_binary_misleads": "'Named on 24 of 60' reads as 40% coverage. Six of those 24 were named exactly once across roughly 39 runs, which is indistinguishable from noise. The binary counts a single lucky run the same as a prompt that names you consistently, and we have none of the latter.",
  "limits": "One brand, one category, 60 prompts, four engines, 10 July to 10 August 2026. The shape of this distribution is ours; the fact that a binary hides it is general.",
  "how_many_zero_runs_to_trust_a_zero": {
    "method": "If a prompt's true rate is p, the chance of n consecutive runs all returning nothing is (1-p)^n. Solving for n gives the runs needed to rule out a rate above p at a given confidence.",
    "table": [
      {
        "rule_out_true_rate_above_pct": 20,
        "confidence_pct": 90,
        "consecutive_zero_runs_needed": 11
      },
      {
        "rule_out_true_rate_above_pct": 20,
        "confidence_pct": 95,
        "consecutive_zero_runs_needed": 14
      },
      {
        "rule_out_true_rate_above_pct": 10,
        "confidence_pct": 90,
        "consecutive_zero_runs_needed": 22
      },
      {
        "rule_out_true_rate_above_pct": 10,
        "confidence_pct": 95,
        "consecutive_zero_runs_needed": 29
      },
      {
        "rule_out_true_rate_above_pct": 5,
        "confidence_pct": 90,
        "consecutive_zero_runs_needed": 45
      },
      {
        "rule_out_true_rate_above_pct": 5,
        "confidence_pct": 95,
        "consecutive_zero_runs_needed": 59
      }
    ],
    "our_own_zeros": {
      "runs_each": 39,
      "confidence_true_rate_below_10pct": 98.4,
      "confidence_true_rate_below_5pct": 86.5
    },
    "correction": "An earlier version of the article recommended five to seven runs before treating a zero as real. That figure was borrowed from a different question - how fast a RATE estimate converges on prompts already known to vary - and does not apply to confirming an absence. Seven zero runs against a true rate of 10% happen 47.8% of the time by chance."
  },
  "by_engine": {
    "note": "Prompts where the brand was never named, per engine, out of the same 60.",
    "engines": [
      {
        "engine": "ChatGPT",
        "prompts": 60,
        "never_named": 46,
        "named": 14,
        "named_pct": 23.3
      },
      {
        "engine": "Google AI Overviews",
        "prompts": 60,
        "never_named": 47,
        "named": 13,
        "named_pct": 21.7
      },
      {
        "engine": "Perplexity",
        "prompts": 60,
        "never_named": 49,
        "named": 11,
        "named_pct": 18.3
      },
      {
        "engine": "Gemini",
        "prompts": 60,
        "never_named": 59,
        "named": 1,
        "named_pct": 1.7
      }
    ],
    "why_it_matters": "36 of 60 are invisible on ALL four engines, which is the intersection of the four never-named sets, but 46 of 60 are invisible on ChatGPT alone. An intersection is necessarily no larger than any single set, so a single-engine check surfaces more gaps than the combined view, and the engine you check changes which prompts look like gaps."
  },
  "prompt_intent": "All 60 are buying-intent: each names or seeks a tool, product or vendor rather than asking a definitional or how-to question."
}