97%of records link to their source
26duplicate records removed
48unconfirmed records held back
The 26 duplicates are part of the 48 held-back records, not in addition to them — 48 records total sit outside the browsable pages.
How this base is checked
Every figure is transcribed from a primary paper or a peer-reviewed meta-analysis and checked against the
source. Numbers that could not be independently confirmed are marked on the record rather than presented as
settled. In July 2026 the full base was audited page-by-page against the underlying data, including an adversarial
review whose job was to try to break our own claims. The categories below are what that produced.
July 2026 audit — what changed
One paper, one count (July 9). The dataset intentionally splits some papers into
several outcome-specific records (a trial that measured both PTSD and depression appears once under each). A
second audit pass found that when several of those records carried the same condition tag, that
condition's headline counted one paper more than once. Study, trial, and participant counts are now computed
per unique paper: PTSD's randomized-trial count corrected from 35 to 27, anxiety's from 55 to 54, depression's
from 36 to 34, and the site-wide totals from 467 studies / 124 RCTs to 446 unique publications / 107 RCTs.
Duplicate records (the same trial catalogued twice) were also merged wherever found — a college-depression
RCT entered twice, a postmenopausal-depression trial published as both a journal article and a conference
abstract, a Korean students' review catalogued three times, and more — bringing the duplicate tally to 26.
Smaller numbers, honestly counted.
Effect sizes transcribed from source (July 9). Most records listed a study but no
comparable effect size. We began reading the primary papers to add the between-group effect size (tapping vs. a
comparison group) for each trial's main outcome, computing it from the paper's own reported means and standard
deviations where it wasn't stated directly, and leaving it blank wherever the paper reported only a within-group
change, only a p-value, or sat behind a paywall we couldn't verify through. Nothing was estimated or carried over
from a secondary summary. During this work a mis-entered DOI was also corrected, and a childbirth-fear trial whose
figures had been accidentally copied from a different study was caught and flagged for re-checking.
"Randomized" checked against the methods (July 9). We re-read the methods section of
studies the dataset had labelled as randomized controlled trials and reclassified the ones that were not:
several were quasi-experimental (a control group, but participants were not randomly allocated to it), and a
handful were single-arm studies with no comparison group at all. A couple of these are large sources — a
"randomized" label was corrected to "outcome study" where a paper described a one-group before-and-after design
or two cohorts that both received tapping. We also removed a registered trial protocol that has no results yet.
The strict site-wide randomized-controlled-trial count moved from 112 to 108 as a result, and several
per-condition trial counts came down accordingly. The studies are all still here and still counted as studies —
they're just labelled for what they actually are.
Geography corrected (July 9). The explore page said 53% of studies were run outside
the United States; the real figure from the dataset is 67%. The "43 countries" count also mixed two spellings
of one country and several multi-country entries — the corrected count of distinct countries is 39.
Real-world app numbers re-verified (July 9). The "in the app" session counts on the
condition pages were re-checked against the app's live public catalog. Several were wrong in both directions
— weight & food cravings said 3 sessions when the app has 27; phobias said 13 when the app has 36;
depression said 34 when only about 14 are on-topic; cancer & serious illness said 45 where an honest count
of genuinely serious-illness sessions is about 28. All counts and descriptions were corrected to match the
actual catalog.
Labels tightened (July 9). The "APA efficacy criteria" tile is now shown only on
clinical psychological conditions, where the American Psychological Association's empirically-supported-treatment
framework actually applies — not on weight, athletic performance, biology/mechanisms, or catch-all categories.
A superseded pre-correction meta-analysis value was removed from the mechanisms chart in favor of its published
corrigendum; two uncontrolled PTSD-outcome data points were removed from the anxiety chart, whose caption
promises controlled comparisons; and two "typical effect" tiles that rested on a single or off-topic value were
corrected (pain now shows "—" until a pain-outcome comparison effect is available).
Inflated effect sizes, corrected. Several effect sizes had been carried over from
secondary sources as if they were a treatment's advantage over a control group, when the primary paper actually
reported a within-group before-and-after change (a larger, less meaningful number). Where the primary paper
published a correction (a corrigendum), we show the corrected figure and note that the original was revised.
Duplicate studies merged. 21 records turned out to be the same study catalogued
twice — often once from a direct source and once from a meta-analysis table, sometimes with slightly different
numbers. We kept the best-sourced version of each and removed the rest so no single study is counted more than
once. The removed records remain in the downloadable dataset, tagged as duplicates, for full transparency.
An out-of-scope study removed (July 9). One record turned out to be a distant
biofield ("energy healing") trial with no tapping or Emotional Freedom Techniques component — it had been
mis-filed as an EFT study. It has been removed from every browsable page and flagged in the downloadable
dataset as out of scope. This directory is for EFT/tapping and Thought Field Therapy only.
Study types labelled honestly. A meta-analysis is not a study; a review is not a
trial; an n=1 case report is not a randomized controlled trial. We corrected records whose design label
overstated what was actually done, and the headline count now reads "studies and reviews" rather than implying
every entry is a primary study.
Claims matched to what was measured. Where a page credited a hard physiological
outcome (like blood pressure) that a study didn't actually find significant, or a pooled effect was described as
covering a whole condition when it only measured one symptom, we corrected the text to match the real result.
An implausible outlier removed, a superseded analysis retired (July 9). A programmatic sweep
of the whole dataset flagged the rest. One PTSD record carried a derived effect size of d=8.07 — far above the
d=3.0 ceiling this site excludes everywhere else — and turned out to be a second copy of a study already listed;
it was merged into the primary record. A dismantling meta-analysis that had since been corrected by a published
corrigendum was retired in favour of the corrected version, so one analysis isn't counted twice. A couple of
journal review articles mis-filed as "dissertations" were relabelled. And, at the display level, a paper split
into several outcome records can no longer appear as more than one card on the same condition page.
Unconfirmed figures held back. 48 records that could not be confirmed as
genuine tapping studies, or whose figures we could not verify, are kept out of the browsable pages entirely and
flagged in the dataset. No figure on this site was invented or estimated — empty is always preferred over
uncertain.
Sources traced. 97% of records now link directly to their original paper by
DOI or publisher page. The remainder are older dissertations and defunct or regional journals with no stable
online record; those are labelled as catalogued from a peer-reviewed index rather than given a broken link.
Standing commitments
Null results and criticisms stay in. Corrections are logged here rather than quietly edited away. Funding by
The Tapping Solution is disclosed on every relevant page and on the governance page.
Where a study was authored by a contributor or funded by the company, that is notable in the record. The aim is a
resource a skeptic can trust — which means showing our work, including the parts we got wrong first.
See also: how studies are included and
graded · governance & independence · report an
error or request the data.