Can THC-Rich Cannabis Increase False Word Recognition When a THC–CBD Strain Does Not? A 116-Person Study

Short answer: In a naturalistic study of 116 regular cannabis users, consuming a THC-dominant flower was associated with poorer discrimination between previously studied and new words, more false recognition of new words, a more liberal recognition response criterion, and slower responses compared with the same participants’ non-use session. The THC–CBD flower group did not show a statistically significant change in discrimination or false alarms, although its response criterion also changed. These laboratory word-recognition results do not measure the truth of an eyewitness account, establish a particular person’s impairment, or validate a blood-THC threshold.

The original primary research is Paulich and colleagues, Naturalistic investigation of cannabis strains varying in THC and CBD ratios and verbal recognition memory, published January 6, 2026, in Frontiers in Psychology, DOI 10.3389/fpsyg.2025.1685412. The interpretation below is limited to that study’s reported design, measured outcomes, and uncertainties.

What question did the investigators actually test?

The investigators asked whether the acute effect of cannabis consumption on recognizing words differed by the THC-to-CBD composition of the cannabis flower. This is narrower than asking whether cannabis always harms memory. Recognition requires deciding whether a presented word was on an earlier study list. A participant can correctly recognize an old word (a hit), incorrectly call a new word old (a false alarm), or make either of the corresponding opposite responses. The experiment examined these response patterns, not autobiographical memory or actual event witnessing.

Its distinguishing feature was the comparison among three commercial flower products: a CBD-dominant, low-THC product, a THC-dominant product with little CBD, and a product containing appreciable amounts of both THC and CBD. The question is especially relevant when a forensic opinion overextends a positive cannabinoid result into a claim about memory accuracy or a statement’s veracity.

Who participated, and how were comparison conditions arranged?

The final analyzed sample comprised 116 adults who used cannabis regularly, with 58 women and 58 men. Participants were randomized to one of three strain assignments: 38 to CBD-dominant flower, 40 to THC-dominant flower, and 38 to a roughly one-to-one THC–CBD flower. The average age was 27.65 years, and average reported cannabis use in the preceding 30 days was approximately 19–21 days depending on assignment.

Each participant experienced a cannabis-use laboratory session and a separate non-cannabis-use session, with the sequence varied. Thus the experimental condition comparison used repeated measurements within people, while the strain comparison was between different people. There were 116 people, not 232 independent participants. Session order, sex, strain, and self-reported frequency of cannabis use were addressed in the researchers’ statistical models.

People used their assigned product at a private residence for approximately 15 minutes, in an amount they chose, and then returned to a mobile laboratory for testing. The non-use session involved a non-cannabis activity. Participants were aware of the commercial products; the design did not include a blinded cannabis placebo. Sessions were separated by several days. Self-selected administration, frequent cannabis use, and product awareness matter when interpreting any apparent cannabinoid interaction.

What THC and CBD products were administered?

The analyzed products were labeled approximately <1% THC and 17.4% CBD for CBD-dominant flower; 12.5% THC and <1% CBD for THC-dominant flower; and 8.2% THC and 6.5% CBD for the mixed THC–CBD flower. Product potency was assessed under the legal-market laboratory-testing framework described by the investigators. The amount of plant material consumed did not differ significantly among groups: reported averages were 381.16 mg, 358.40 mg, and 381.27 mg, respectively; F(2,112) = 0.16, p = 0.85.

By multiplying product weight by labeled potency, researchers estimated contained THC amounts of roughly 3.81 mg, 44.80 mg, and 31.26 mg across those groups. Their corresponding contained CBD estimates were 66.32 mg, 3.58 mg, and 24.78 mg. These figures are estimates of material-associated cannabinoid content. They are not measured quantities absorbed, delivered to the lungs, retained after combustion, or present in an individual’s blood. The method cannot establish a standardized THC dose across participants.

What did the verbal-recognition task measure?

Participants studied sets of words and were then presented with previously studied (“old”) and previously unseen (“new”) words. The design included seven study lists and recognition testing with old/new probes, with intervening attention tasks. A response to a new word counted as a false alarm if the participant called it old. An accurate “old” judgment of a truly studied word counted as a hit. Investigators examined discrimination sensitivity (d′), response criterion (C), false alarms, hits, and response latency.

Discrimination and response bias are not identical. The d′ index describes how effectively participants distinguish studied from new words; a lower number reflects weaker separation. Response criterion captures a tendency to answer “old” or “new” more liberally or conservatively. A more liberal criterion may generate false alarms even when the person’s ability to distinguish categories does not change in the same way. The two should not be conflated with a witness’s honesty or the accuracy of an entire narrative.

The researchers additionally tested free recall approximately two hours after cannabis use, asking participants to retrieve words without the same recognition cues. Its results must be reported separately from recognition because these are different cognitive endpoints.

What were the observed raw recognition means?

The original paper’s Table 2 provides the following descriptive values. They are group means, not individual diagnostic thresholds, and the inferential models additionally adjusted for relevant covariates. Raw differences should not be substituted for the adjusted simple-effect estimates reported in the Results.

For the THC-dominant group, the raw false-alarm proportion was 0.28 after use and 0.19 during non-use. That difference means, on average, a greater fraction of new test words were incorrectly treated as studied words in the use session. It is not a 28% probability of false eyewitness identification and cannot predict whether a named witness is mistaken.

Which recognition comparisons were statistically significant?

The adjusted analysis found a strain-by-session interaction for discrimination, F(2,103) = 4.70, p = 0.01, partial η² = 0.08. Follow-up tests attributed a significant discrimination decline to the THC-dominant group: the estimated non-use-minus-use difference was 0.42 (SE 0.134; p = 0.002). The corresponding within-group effects were not statistically significant for either CBD-dominant or mixed THC–CBD flower.

For false alarms, the strain-by-session interaction was F(2,103) = 7.71, p < 0.001, partial η² = 0.13. The THC-dominant group’s adjusted false-alarm increase was approximately 0.08 (SE 0.02; p < 0.001). The other two groups did not produce significant within-group false-alarm effects. Importantly, the analogous interaction for hit rate was not significant: an overall “memory accuracy fell on every metric” statement would be inconsistent with the reported results.

Response criterion showed a separate interaction, F(2,103) = 3.82, p = 0.025, partial η² = 0.07. Both THC-dominant (adjusted difference 0.12, SE 0.05, p = 0.02) and mixed THC–CBD flower (difference 0.15, SE 0.05, p = 0.002) produced more liberal criteria after use. The mixed group’s criterion shift is an important positive result even though its discrimination and false-alarm simple effects were not statistically significant. It prevents an oversimplified assertion that adding CBD removed every memory-related cannabis effect.

Did response speed and free recall change in the same way?

Not entirely. Recognition response time yielded a significant strain-by-condition interaction, F(2,97) = 3.25, p = 0.04, partial η² = 0.06. The THC-dominant group responded more slowly following use in the adjusted follow-up analysis (difference 0.11 in the original analysis’ reaction-time scale, SE 0.04, p = 0.004). Other strain groups’ corresponding changes were not statistically significant. Do not convert this value to milliseconds without confirming the source scale.

By contrast, the interaction for the secondary free-recall outcome was F(2,97) = 0.30, p = 0.75, partial η² = 0.01. This null result is not evidence that all dimensions of memory were normal or unchanged in every person. It says that this particular analysis did not demonstrate differing use-versus-non-use effects among the three strain groups on the reported free-recall measure.

Did the investigators measure blood THC?

Yes. The investigators collected blood around experimental sessions and reported mean concentrations of parent THC, CBD, THC-COOH, and 11-hydroxy-THC in ng/mL. However, the accessible main primary paper does not sufficiently specify the precise analytical blood fraction and instrument to justify relabeling those values as a forensic venous whole-blood GC–MS or LC–MS/MS measurement. “Blood cannabinoid concentration” must retain its source’s analytical qualification until the laboratory protocol is independently verified.

These group means after cannabis use verify that the flower assignments produced different cannabinoid exposures. They do not establish a dose–effect curve, a validated concentration cutoff for recognition errors, or a precise time of last use for an individual. THC-COOH is a metabolite and is not interchangeable with active parent THC, even when both are reported in the same concentration unit.

The investigators reported similar mean post-use THC and self-rated intoxication between the two THC-containing groups, but substantially different CBD concentrations. This supports the intended exposure contrast; it does not independently isolate CBD as a causal antidote to THC. The products also differed in THC potency, the quantity of plant constituents, and unblinded consumption patterns. Group mean matching does not imply every participant received the same absorbed THC amount.

Can the results prove CBD protects a person’s memory from THC?

No. The mixed THC–CBD group did not show a statistically significant deterioration on the primary discrimination or false-alarm comparisons, while the THC-dominant group did. That pattern is consistent with a possible moderating role for CBD and justified the authors’ scientific interpretation. But a nonsignificant within-group effect is not proof of equivalence, and a positive comparison between differently formulated flower products is not the same as administering precisely matched THC with randomized CBD supplementation under blinding.

To establish a specific pharmacological protective effect would require more controlled dosing, clearer measurement of absorbed exposure, a blinded placebo or comparator structure, adequately powered effect estimates, and replication across user histories and tasks. These data provide evidence about the performance of groups under these conditions, not an individual patient’s CBD rescue or an asserted forensic threshold.

What does “false recognition” mean in a legal setting?

In this experiment a false recognition was a concrete laboratory outcome: a participant selected “old” for a word the investigators knew had never appeared on the study list. Researchers could score the answer against a known ground truth. A statement by an eyewitness concerns complex sensory exposure, divided attention, event meaning, delay, questioning, stress, confidence, and potential contamination. A word-list false alarm is not the same construct as fabricated testimony, conscious deception, mistaken identification, or every kind of eyewitness error.

The study could support a carefully qualified general proposition that cannabis composition and acute use can affect selected recognition decisions in frequent users. It cannot provide a defensible numerical probability that a particular eyewitness’s account was wrong or that a person was truthful or untruthful. To evaluate a real event, the relevant perception conditions, timeline, contemporaneous behavior, available toxicological samples, and alternative explanations remain separate questions.

Could a forensic blood report establish this participant’s memory impairment?

Not from this paper. The source shows between-group differences in average measured cannabinoid concentrations and group-level recognition performance. It does not develop or validate a blood-THC classification model for an individual recognition test, estimate case-specific sensitivity or specificity, or prove that a single concentration causes a specified false memory. Repeated-user baseline samples contained positive cannabinoids even outside the experimental use condition, emphasizing that concentration and precisely timed function are different measurements.

Specimen matrix matters. Serum, plasma, and whole blood may have different analyte concentrations and collection-dependent interpretation; metabolites may reflect earlier exposure. The source’s generic “blood” category should not be silently imported into a forensic whole-blood test report. Likewise, product milligrams and subjective intoxication are neither laboratory blood concentrations nor interchangeable measures of cognitive function.

What limitations most constrain courtroom interpretation?

All participants were adult regular users in one naturalistic study, not random members of the driving or eyewitness population. The flower groups differed between persons; sessions crossed use and non-use within persons. Cannabis amount was self-selected, and participants knew the market product used. The design assessed recognition words in a mobile-laboratory environment, not authentic witnessed events or driving. The effect estimates reflect particular tasks, timing, and analytic adjustments. The raw means, model differences, and nonsignificant results must be kept distinct.

The evidence also depends on the specific product profiles rather than a simple binary THC-present/CBD-present label. Even a statistically significant result describes an estimated sample effect, not a universal law of memory. Other determinants of performance—prior cannabis use, sleep, illness, medications, task practice, cue salience, and individual variability—remain relevant. Null contrasts especially should not be treated as conclusive demonstrations of no effect.

What should an evidence-bound forensic report actually say?

A source-grounded report may state that, among 116 regular users assigned to three cannabis flower types, a THC-dominant group displayed poorer recognition discrimination and more false alarms following use relative to a non-use session, while a THC–CBD group did not show a significant deterioration on those specific metrics. It should add that both THC-containing groups showed a more liberal response criterion, that hit-rate and free-recall interactions were not significant, and that researchers did not derive a legal impairment threshold or validate witness accuracy.

Such an account is scientifically useful precisely because it respects the limits of the measured endpoint. It avoids asserting that all THC exposure causes false testimony, that CBD neutralizes every cognitive effect, that positive cannabinoids prove an unreliable witness, or that a study-group blood mean identifies a particular person’s impairment.

Frequently asked questions

1. Does the study mean cannabis users always remember events incorrectly?

No. It involved a defined word-recognition task with different effects among product groups. The results cannot establish an error rate for real-life memories or all cannabis users.

2. Is a false alarm the same as lying under oath?

No. A false alarm was an incorrect “old” selection for a new test word, not a measure of intent, deception, honesty, or eyewitness truth.

3. Was the study a randomized trial of CBD added to an identical THC dose?

No. People were randomized to three commercial flower profiles, but self-administered unblinded product amounts and the formulations and THC content differed. The observed contrast does not conclusively isolate CBD.

4. Did the mixed THC–CBD strain have no cognitive effect?

No such blanket conclusion is justified. Its discrimination and false-alarm changes were not significant, but its response criterion became more liberal after use, a statistically significant result.

5. Could the published blood values be read as forensic whole-blood THC cutoffs?

No. They are reported study-group blood-cannabinoid means. The accessible original does not resolve the specific blood fraction or validate any concentration cutoff for individual recognition impairment.

6. Did the THC-dominant group forget every studied word?

No. Group-level discrimination declined and false alarms increased, but the hit-rate interaction was not significant. Memory performance has several noninterchangeable measurements.

7. Could this article alone establish intoxication, driving impairment, or witness inaccuracy in a lawsuit?

No. It evaluated laboratory recognition in regular users, not real driving or the accuracy of an actual witness statement. Case-specific toxicological methods, behavior, timing, and competing explanations would require separate evaluation.

Source and related scientific context

Original full primary: Paulich KN, Place C, Giordano G, Carpenter WB, Curran T, Bidwell LC. Naturalistic investigation of cannabis strains varying in THC and CBD ratios and verbal recognition memory. Frontiers in Psychology (2026); DOI 10.3389/fpsyg.2025.1685412. For broader methodology, see the same site’s published resources on biological matrices in toxicology, specimen collection and limitations, and confirmatory drug testing. These contextual links do not substitute for the original study or establish a case-specific conclusion.