Living research · pre-registered · stable URL

Does buying directory listings change whether AI engines cite you?

By Reviewed & updated Measurement-first: figures are median share-of-model with 95% confidence intervals. How we measure.

The only two studies that exist are vendor-published and flatly contradict each other, and nobody has run a controlled before/after and published the method. We are — on ourselves, in public, pre-registered before we knew the answer. We publish the null result the same as a win.

We sell this. We measured ourselves anyway. Clear Cited runs the exact service we sell on clearcited.com and publishes the numbers — the flat months included. Here's the method, the confidence intervals, the pre-registration hash, and the raw data. Check us. How we measure →
Measuring. Pre-registered. Measuring now. Nothing has moved yet, because there is nothing to move yet — the baseline is being sealed. This page publishes the answer the moment the first live cycle lands — flat, negative or positive, all the same.

The pre-registration

The hypothesis, the frozen prompt set, the keyword set, the metric set and the cadence were sealed and hash-chained before cycle 000 — commitment hash e692bc9506082090. We cannot move the goalposts after the fact; that is the entire difference between this and a vendor study.

The frozen buyer-prompt set

The method

The method (check us)

  • Cadence: every 30 days from T0, forever — flat cycles published too.
  • Reproducible: median of ≥10 runs per engine across ChatGPT, Perplexity, Claude, Gemini, and Grok (never below a 5-run floor), with metric-matched 95% confidence intervals — presence via Wilson, share via bootstrap. AI Overviews measured as a separate stream, never summed in.
  • Pre-registered: the hypothesis, the frozen prompt set, the frozen keyword set and the metric set were sealed and hash-chained before cycle 000 — commitment hash e692bc9506082090.
  • Honest limits: n=1, no control, confounded, tiny samples disclosed at every stage. Read directional, not causal.
  • Raw data: CSV · JSON — ungated, no email wall. Full methodology →

The directory submission is gated behind the sealed T0 baseline (W320/W321) and has not run yet; when it does it is logged in the ledger below and the before/after reads against it.

Share of model over time

Share of model — clearcited.com (before/after the directory push) — baseline being measuredConnect the source or add a manual reading - we never invent a number.
Measured share-of-model over time — the same numbers as the chart above, as a table for machine extraction and screen readers.
CycleAs ofShare of model95% CIStatus
No measured cycle yet — the baseline is being measured.

Which directories does AI actually cite?

The commercially useful question, and the one our clients pay us to answer: of the directories the tactic buys, which does an AI engine ever actually reach for in our category? Measured, not asserted. "Not yet observed" is the honest answer we publish — as loudly as a hit.

Directory citation scoreboard — for each directory in the tactic, whether an AI engine has been MEASURED citing it behind an answer about us. 'Not yet observed' is the honest null, published as loudly as a hit.
DirectoryEver cited (measured)Cycles
g2.comnot yet observed
capterra.comnot yet observed
getapp.comnot yet observed
softwareadvice.comnot yet observed
trustradius.comnot yet observed
producthunt.comnot yet observed
alternativeto.netnot yet observed
saashub.comnot yet observed
sourceforge.netnot yet observed
slant.conot yet observed
crozdesk.comnot yet observed
goodfirms.conot yet observed
clutch.conot yet observed
llmstxt.sitenot yet observed

The published intervention ledger

Every dated material action in the measurement window — so a reader can see the confounders rather than take our word that we accounted for them. n=1, no control, confounded; the ledger is how you read the series honestly.

The published intervention ledger — every dated material action on clearcited.com in the measurement window. Directory actions are flagged; the confounders are shown, not hidden.
DateSurfaceActionDirectory?
2026-06-14siteSite launched (T0 anchor for the whole-business baseline)

Limitations

Limitations — in the body, not a footnote. Stating these makes the study more credible, not less.
  • n=1 — a single subject (ourselves); results do not generalize by themselves.
  • No control group — there is no unexposed clearcited.com to compare against.
  • Confounded — many interventions run at once; the ledger records them but cannot isolate their effects.
  • Tiny samples at every funnel stage — sample size is disclosed at each stage and read as directional.
  • AI-search stochasticity — roughly 34–42% run-to-run source overlap; mitigated by median-of->=10 + CIs, not eliminated.
  • Forward-only instruments — analytics/GSC/rank-tracking series start at their install/verify date, disclosed per series.
  • Directory-specific confounding — the directory submission runs in the same window as content, entity and earned-citation work; the ledger below records every one so a reader can see the confounders rather than take our word we accounted for them.
  • One tactic, one subject — a finding here is about the directory tactic on clearcited.com, not a law about directories; it does not disparage any supplier who executed the submission as ordered.

Dispute the data · challenge the method

Attack this. Inviting people to attack the method is the strongest credibility signal available to us. The raw data is ungated below — download it, re-run it, tell us where we are wrong. Email [email protected] with a dispute or a method challenge; granted corrections are logged in the corrections log, publicly, with a date.

This is a flow, not a slogan: the corrections log below is append-only and every entry is dated.

Corrections log

Append-only. Empty until the first granted correction — which is itself the honest state, published.

  1. 2026-07-13 — Study pre-registered and hash-chained commitment hash e692bc9506082090 — hypothesis, frozen prompt set, keyword set, metric set and cadence sealed before cycle 000

The raw data (ungated)

No email wall. Download the full series and the scoreboard — CSV · JSON. Licensed CC BY 4.0. Reuse it with attribution; if you find we are wrong, we will say so here.

Suggested citation. Clear Cited, "Does buying directory listings change whether AI engines cite you?", https://clearcited.com/research/directories-and-ai-citation/. Licensed CC BY 4.0.