External Sources: CRO, Paid Ads, Cannabis and Hemp Ecommerce
What is worth acquiring from outside, what is not, and what only we can build.
Compiled 2026-08-11. Every source below was checked live on that date, not recalled. Scope: TFC Marketing accounts (TGH Houston, Cali Extrax, Wonder Fungi). Nothing here is brand-specific, so nothing here breaks per-brand isolation.
0. How to read this file
Same evidence discipline as dev\klaviyo-cannabis-strategist\knowledge.md. It maps one to one
onto the (Client-Confirmed) / (Research-Indicated) tags in the Rise template set:
- VERIFIED means the publisher's own page was read directly on the date shown. Prices, methodology statements and document inventories carry this only when they came off the source itself. Equivalent to (Client-Confirmed).
- INHERITED means a search summary or a secondary write-up said it. Useful for orientation, not quotable to a client or a colleague on its own. Equivalent to (Research-Indicated).
- JUDGEMENT means this is an assessment, not a fact. It is arguable and it is Marita's to overrule. Assessments do not carry a recheck date because they do not rot, they get revised.
- RECHECK-BY goes on anything that rots. Prices and free-tier boundaries rot fastest.
A publisher's claim about its own rigour is a marketing claim until someone reads the method. Where a source states its own sample size or hours of research, that is recorded as their claim, not as an independent finding.
1. The one distinction that decides everything below
Marita's instinct on Baymard is right, and it generalises. There are four tiers of evidence in marketing literature, and almost every argument about "best practice" collapses once you sort a source into one of them.
Tier A. Moderated observation. A recruited participant is watched, on a real site, trying to do a real task, and the failure is recorded with the reason. Baymard and Nielsen Norman Group are the only two operating at this scale in ecommerce. The output is causal: it tells you why something fails, which is the only kind of finding an agent can generalise from when writing copy for a site it has never tested.
Tier B. Controlled experiment with published effect sizes. Real A/B tests with sample sizes and confidence stated. GoodUI is the only public library of these. The output is outcome data: it tells you whether a change moved money, on someone else's site, once. It does not tell you why, and it does not transfer cleanly.
Tier C. Population telemetry. Real measurements across millions of real sites, with no intervention. HTTP Archive and the Chrome UX Report. The output is descriptive: what the web actually does, which is the right way to settle "is this normal" arguments and the wrong way to settle "does this work" arguments.
Tier D. Install-base aggregates and lead magnets. A tool vendor reports averages across its own customers, or an agency publishes a guide. The sample is defined by who bought the tool or who the agency wants to sell to, the analysis is unpublished, and the document exists to generate a lead. Almost all cannabis marketing literature is Tier D. Almost all ESP and CRO platform "benchmark reports" are Tier D.
The rule this file applies: Tier A and B findings can enter a playbook as guidance. Tier C can settle a factual dispute. Tier D can only ever be a pointer to a primary source, never a citation. JUDGEMENT.
2. CRO
Ranked
1. Nielsen Norman Group, ecommerce report series. Buy this first.
What it is. Thirteen volumes on ecommerce UX. The full series carries 1,037 recommendations and 3,100+ screenshots. Individual volumes include Product Pages (428 pages, 108 recommendations, 5th edition), Search and faceted search (444 pages, 139 guidelines), Trust and Credibility (174 pages, 53 recommendations), Selling Strategies (331 pages, 102 recommendations), plus Shopping Carts and Checkout, Customer Service, Transactional Email and Notifications. VERIFIED 2026-08-11, nngroup.com/reports/topic/e-commerce and the Product Pages report page, read directly.
Who publishes it. Nielsen Norman Group, founded by Jakob Nielsen and Don Norman. An independent research and training firm with no software product to sell.
Free or paid. Paid, but priced per report, not as a subscription: $98 individual license, $198 group license per volume. NN/g also publishes a set of genuinely free reports at nngroup.com/reports/free and a large free article library. VERIFIED 2026-08-11 from the Product Pages report page. RECHECK-BY 2026-11-11.
Why it is credible. Tier A. Five separate rounds of ecommerce studies using one-to-one moderated usability testing, a diary-based longitudinal study, eyetracking, and expert review. Representative users tested 350+ ecommerce sites across the United States, United Kingdom, Denmark, India and China. The method is stated on the report page rather than in a press release. VERIFIED 2026-08-11.
What it improves here. This is the single best value-per-dollar acquisition on the list. Four relevant volumes cost roughly $400 once, no renewal, and they can be read into a CRO playbook that the specialist agents then read before writing. The Transactional Email and Notifications volume is directly in the email lane, not just the CRO lane, which no other source on this list covers with Tier A evidence.
Caveat, stated honestly. No publication date is shown on the report pages, only an edition number. Some of the underlying studies are old. Ecommerce interaction patterns move slowly, so this matters less than it would in a fast-moving field, but a finding should be sanity-checked against a live site before it becomes a rule. JUDGEMENT.
2. Baymard Institute. The strongest source, and the most expensive.
What it is. A continuously updated database of ecommerce UX guidelines with performance scores and best-practice examples, plus a benchmark database of real checkout flows. Cart and checkout research alone runs to 110+ findings across 17 topics and 611 pages. Guideline and page counts VERIFIED 2026-08-11 from baymard.com/pricing; the 611-page checkout figure is INHERITED 2026-08-11 from secondary write-ups.
Who publishes it. Baymard Institute, an independent research firm. No software product.
Free or paid. Paid, annual, with a real free tier.
- Free plan, $0. A limited set of guidelines and selected page design examples.
- Foundation, $200/mo or $2,400 billed annually. 150 foundational guidelines covering homepage, navigation, categories, product lists and product pages. 3 users.
- Expansion, $350/mo or $4,200 billed annually. 400+ full-journey guidelines adding cart, checkout, search and account, plus industry-specific insight for 3 industries. 3 users.
- Custom, from $700/mo or $8,400 billed annually. 700+ guidelines adding accessibility and loyalty, plus recurring consultations. All VERIFIED 2026-08-11, baymard.com/pricing read directly. RECHECK-BY 2026-11-11.
Separately, Baymard publishes free, citable statistics pages. The cart abandonment page is a meta-analysis of 50 studies giving an average abandonment rate of 70.22%, with a documented range of 55% to 84.27% across those studies, and a breakdown of stated abandonment reasons (39% surprised by extra fees, 17% checkout too long or complicated). Their checkout benchmark finds the average US checkout shows 23.48 form elements by default against an ideal of 12 to 14. INHERITED 2026-08-11 from the search result summary of baymard.com/lists/cart-abandonment-rate; the page itself was not opened, so treat the numbers as orientation until read. RECHECK-BY 2026-11-11.
Why it is credible, and what the difference actually is. This is the point worth writing down, because it is the argument that decides which sources get into a playbook at all.
Baymard's findings come from moderated usability testing: a recruited participant, a real task, a real site, watched. The finding is the observed failure and its cause. A tool vendor's "benchmark report" comes from aggregating its own install base: the sample is whoever bought the tool, the analysis is unpublished, and the report exists to sell the tool. The two are not the same kind of claim and should never sit in the same citation list.
The practical consequence: a Baymard finding tells you why a shopper abandoned, so it generalises to a site nobody tested. A vendor aggregate tells you what the average was among that vendor's customers, which generalises to nothing. JUDGEMENT.
Their own claim about scale. Baymard's site states 200,000+ hours of UX research. Widely circulating third-party write-ups cite 28,000+ hours, which appears to be a much older figure still being copied. Both are Baymard's own numbers at different points in time, neither is independently audited, and the discrepancy is a good illustration of how a marketing figure propagates. Do not quote either to a client. 200,000+ VERIFIED 2026-08-11 as appearing on baymard.com. 28,000+ INHERITED 2026-08-11 from a third-party page. Neither is verified as true.
What it improves here. Age-gated hemp storefronts carry every standard checkout defect plus an extra interstitial before anyone sees a product. Baymard is the only source with enough depth on cart and checkout to write a real CRO playbook from. It would give the CRO recommendations that route to Sehrish Iqbal an evidence base instead of an opinion.
Recommendation. Not yet. Start with the free tier plus NN/g. Buy Expansion at $4,200/yr only if CRO becomes a larger share of her billable work than email, because the cart and checkout content is in Expansion and not in Foundation, and Foundation without checkout is the half that matters least for a hemp DTC store. JUDGEMENT.
3. HTTP Archive Web Almanac. Free, and the best free thing in CRO.
What it is. An annual, chaptered report on the real state of the web. The 2025 edition has 16 chapters including a dedicated Ecommerce chapter (platform adoption, Core Web Vitals, Lighthouse metrics, payments, regional differences) and a Performance chapter. Primary analysis is built from July 2025 measurements spanning millions of sites, combining HTTP Archive lab data with Chrome UX Report field data from real Chrome users. INHERITED 2026-08-11 from the almanac.httparchive.org search summary. The chapters are free and open at almanac.httparchive.org/en/2025/ecommerce. RECHECK-BY 2026-11-11 for a 2026 edition.
Who publishes it. HTTP Archive, a community project, with named chapter authors and a published methodology. Google contributes the CrUX data but does not own the report.
Free or paid. Free, no registration.
Why it is credible. Tier C. It is a census, not a survey, and the data collection is automated rather than self-reported. It cannot tell you whether a design change works. It can tell you, definitively, what is normal.
What it improves here. It settles platform and speed arguments with real numbers instead of vendor claims, which is exactly the kind of finding that has to survive being handed to a client's developer. Low effort, zero cost, worth reading once a year.
4. GoodUI. Useful, and the caveats are load-bearing.
What it is. A library of A/B tested UI patterns with published effect sizes. Live counts read off the site: 663+ tests, 141+ patterns, 112+ "Leaks" (observed competitor experiments). Individual tests display total sample size and statistical power at a stated minimum detectable effect, which is unusually honest for this genre. VERIFIED 2026-08-11, goodui.org read directly, for the counts and for the presence of per-test sample size and power figures.
Who publishes it. Jakub Linowski, an independent A/B test designer.
Free or paid. Freemium and the split matters. Free: browse pattern titles, the 75 original UI ideas with descriptions, test descriptions and contributor profiles. The effect sizes are hidden on the free tier (they render as "X.X% Sales"), which is the entire value. VERIFIED 2026-08-11 from goodui.org. Pricing was not visible on the public site and sits behind a "Get Access" gate. Secondary sources report Solo $72/mo ($60/mo billed annually), Team $144/mo ($120 annual), an expert-guided tier around $1,950/mo, and one-off Datastories PDFs at $289. INHERITED 2026-08-11 from a search summary, not confirmed on a live pricing page. RECHECK-BY 2026-11-11.
Why it is credible, with limits. Tier B, and the limits are real. Tests are contributed, so the sample is self-selected. Published experiments skew toward wins, which is publication bias by another name. Many individual tests have small samples (two recent tests carried totals of 3,838 and 9,611 visitors), and a test with a total sample of a few thousand can only detect a large effect. Reading a small-sample win as a rule is exactly the error her own Klaviyo knowledge base already warns about in the A/B testing section.
What it improves here. Best used as a hypothesis generator, never as proof. It is the fastest way to get a specialist agent to propose a test worth running rather than a change worth arguing about.
5. Analytics-Toolkit and "Statistical Methods in Online A/B Testing", Georgi Georgiev.
What it is. A statistics book plus a set of A/B testing calculators, including a sample size and power calculator supporting multiple variants, binomial and continuous metrics such as average revenue per user, and percent-change lift. INHERITED 2026-08-11 from analytics-toolkit.com search summaries.
Who publishes it. Georgi Georgiev, an applied statistician. Reported to have written 100+ articles and papers, won the 2024 Experimentation Thought Leadership Award in Data and Analytics, and to have calculators cited in 400+ scientific papers. INHERITED 2026-08-11. The 400+ papers figure is his own site's claim and is not independently checked here.
Free or paid. Calculators are free to use. The book is paid.
Why it matters here specifically. This is the cheapest fix to a defect that is already documented in her own knowledge base: tests running on accounts where the volume can never reach significance, and automatic winner selection resolving them anyway. A Cali Extrax list of roughly 27 mailable contacts cannot resolve anything, and being able to say so with a number attached is worth more than another CRO guideline.
What it improves here. Turns "this test cannot reach significance" from an assertion into a calculation that survives being challenged by an account manager.
6. MeasuringU, Jeff Sauro. Optional.
What it is. Quantitative UX methodology, the SUPR-Q standardised website quality questionnaire (eight items, four factors: usability, trust, appearance, loyalty, normed against 4,000+ respondents across 100+ websites), and the book Benchmarking the User Experience. INHERITED 2026-08-11 from measuringu.com and the UXPA Journal paper summary.
Free or paid. Mixed. Peer-reviewed papers are free via UXPA Journal. Benchmark data and courses are paid.
Why it is credible. SUPR-Q is published in a peer-reviewed journal, which is a higher evidentiary bar than anything else on this list.
Why it ranks low anyway. It answers "how good is this site, numerically, against a norm", which is not a question anyone at TFC is currently asking. Revisit only if she is ever asked to prove site quality improved over time. JUDGEMENT.
7. Growth.Design. Free, well made, not evidence.
Comic-format product teardowns tied to named psychology principles. Free, large library, genuinely good at giving vocabulary for defending a design decision to a stakeholder. But the principles are asserted and applied post hoc to someone else's shipped product, with no test behind the claim. Read it for language. Do not cite it, and do not let an agent quote it as a reason. JUDGEMENT.
Not worth the time, CRO
- CXL / CXL Institute. Training and agency content, not original research. The case studies circulated under its name (for example the widely repeated Walmart figures) are secondary retellings. Good course, wrong genre for a knowledge base. JUDGEMENT.
- Any "cart abandonment statistics 2026" listicle. The search results for that phrase return a dozen near-identical pages, and tracing them lands back on the same Baymard meta-analysis. Go to Baymard's free list directly and skip the intermediaries. VERIFIED 2026-08-11 by observing exactly this pattern in the results.
- CRO platform benchmark reports (testing tools, heatmap tools, ESPs). Tier D. Their sample is their own customer base and the report is a lead magnet. JUDGEMENT.
3. Paid ads
The category makes this unusual. The major platforms largely refuse this inventory, so general paid-media literature written for Meta and Google buyers describes a machine she cannot use. What transfers is: (a) the primary policy documents, (b) the claims and substantiation law, and (c) measurement methods that do not depend on a platform pixel. Everything else is noise.
Ranked
1. FDA warning letters for cannabis-derived products. Free, primary, and the highest-value acquisition on this entire list.
What it is. FDA maintains a public page, "Warning Letters for Cannabis-Derived Products", listing enforcement letters with the company, the date, and the specific claims cited. The most frequent violation is unapproved new drug claims, with letters citing claims that a product treats cancer, anxiety, pain, arthritis, Parkinson's, Crohn's disease or PTSD. INHERITED 2026-08-11 from the fda.gov search result set. The page exists at fda.gov/news-events/public-health-focus/warning-letters-cannabis-derived-products and should be read directly before anything is built from it. RECHECK-BY 2026-09-10.
Who publishes it. The FDA. Primary regulator, no product to sell.
Free or paid. Free.
Why this ranks first. Her stated problem is copy that misses and gets caught in QA. The claims rule in her knowledge base is currently correct but abstract: "no health, medical or therapeutic claims". That is a principle, and a principle does not stop a generated draft. A warning letter corpus turns the principle into a list of exact sentences the government has already objected to, with a citation attached to each one. That is the form a specialist agent can actually enforce at write time rather than at review time.
Her own knowledge base flags this: item 12 in the unsubstantiated-facts list reads "CBD warning letters are routine. Almost certainly true and easy to substantiate from the FDA's own warning letter database." This closes that item and produces a working artifact in the same pass.
What it improves. Feeds directly into a banned-phrase document (section 5). Also promotes an INHERITED line in the knowledge base to VERIFIED.
2. FTC Health Products Compliance Guidance, plus the Endorsement Guides and the Negative Option Rule.
What it is. Issued 20 December 2022, the first update in nearly 25 years to FTC guidance on advertising claims for health-related products. It expanded scope from dietary supplements to all health-related products. It states that the Commission will enforce a requirement of "competent and reliable scientific evidence" for health-related claims, and that this generally means randomized, controlled human clinical testing by relevant experts. Animal and in vitro studies do not substantiate a health claim without human RCT confirmation. The guidance draws on over 200 cases since the previous version. VERIFIED 2026-08-11 as to content and date, consistent across the FTC's own press release and four independent law firm analyses (Morrison Foerster, Cooley, Wilson Sonsini, Venable). The FTC page itself is at ftc.gov/business-guidance/resources/health-products-compliance-guidance and was not opened directly. RECHECK-BY 2026-11-11.
Adjacent and worth pulling in the same pass:
- FTC Endorsement Guides, which govern testimonials, reviews and creator content. The creator's claim is the brand's claim, which the ads knowledge base already states as a rule without a citation behind it.
- FTC Negative Option Rule, reported to have reached full enforcement in mid-July 2025 and to cover CBD and hemp THC subscription clubs. INHERITED 2026-08-11 from cannabisregulations.ai, a secondary source. Verify against the FTC before relying on it. RECHECK-BY 2026-09-10.
Free or paid. Free.
Why it is credible. Primary. It is the standard she would be measured against.
What it improves. It is the citation behind the claims rule. Right now that rule is defended as "standing category practice", which is a weaker position than naming the document. It also sets the bar for what counts as substantiation if a client ever pushes back on a rejected line.
3. National Advertising Division case decisions, BBB National Programs.
What it is. The advertising industry's self-regulatory body. NAD hears competitor challenges to advertising claims and publishes decisions. Reporting indicates NAD, CARU and the FTC increased scrutiny of hemp-CBD and THC advertising through 2025, specifically on fast-acting, nano and microdose claims, evidence standards, disclosure, and child appeal. NAD has referred at least one CBD advertiser (CannaPure CBD Oil) to the FTC for non-participation. Decision summaries and the CannaPure referral VERIFIED 2026-08-11 as published on bbbprograms.org. The characterisation of a 2025 enforcement expansion into cannabinoids is INHERITED 2026-08-11 from cannabisregulations.ai, a secondary source. RECHECK-BY 2026-09-10.
Free or paid. Summaries are free at bbbprograms.org/media/newsroom/decisions. Full decision text requires a paid Online Archive subscription. Academics and press can request access. VERIFIED 2026-08-11, read directly off the decisions page.
Why it is credible. These are adjudicated disputes with reasoning, not guidance. NAD applies a substantiation standard that in practice runs at or above the FTC floor, because the challenger is a competitor with an incentive to find the weakest claim.
What it improves. The summaries alone are enough to see which claim structures lose. Higher signal per page than any agency guide in the category. Free tier is sufficient.
4. Google Meridian and Meta GeoLift. Free, open source, and probably not usable yet at her spend.
Meridian. Google's open-source Bayesian marketing mix modelling package, Python, on GitHub
and PyPI as google-meridian. Announced March 2024, launched broadly February 2025, and in
February 2026 Google added Scenario Planner, a no-code budget modelling interface that does
not require Python. It supersedes Google's earlier Lightweight MMM.
INHERITED 2026-08-11 from blog.google, PyPI and multiple vendor write-ups; the GitHub repo was
not opened. RECHECK-BY 2026-11-11.
GeoLift. Meta's open-source geo-experiment library, R, MIT licensed, based on synthetic control methods, actively maintained at github.com/facebookincubator/GeoLift. Works on aggregated market-level data, so it survives signal loss. INHERITED 2026-08-11.
The honest caveat, and it is decisive. A 2026 head-to-head simulation study from Recast reports GeoLift holding false positives near nominal (3 to 5%) but missing 89 to 96% of real effects compared with other open-source geo tools (CausalPy, Google Matched Markets, CausalImpact). INHERITED 2026-08-11 from research.getrecast.com, and note that Recast sells a competing commercial MMM product, so the result is not disinterested.
More importantly: her ads knowledge base records roughly $1,180 of spend in a measured week on TGH. Neither MMM nor a geo holdout produces a readable signal at that budget. Her own knowledge base already says this ("a holdout at $400 a month will not produce a signal, and saying so is more useful than running it"), and that position is correct.
What it improves. Nothing this quarter. Acquire the knowledge, do not acquire the tool. The value is being able to say precisely why incrementality testing is not available at this spend, which is a better answer than promising one and failing to deliver it. JUDGEMENT.
5. IPA Databank, Binet and Field. Good argument, biased sample.
What it is. The Long and the Short of It, published 2013 for the Institute of Practitioners in Advertising, analysing 996 campaigns entered into the IPA Effectiveness Awards between 1980 and 2010. Source of the 60/40 brand-building versus sales-activation split. INHERITED 2026-08-11 from ipa.co.uk and multiple secondary summaries.
Free or paid. The summary presentation PDF is free on ipa.co.uk. Full Databank access is an IPA member benefit.
Why it is credible, and where it is not. It is the largest published effectiveness dataset in existence. But the sample is awards entries, which are self-selected, self-reported and submitted by agencies with an interest in the result. It is evidence about campaigns that agencies chose to enter and write up, not about campaigns in general. Say that out loud whenever the 60/40 number is used. JUDGEMENT, and it is the standard critique, not a novel one.
What it improves. Gives her a defensible answer when a client wants every dollar on bottom-funnel activation. In a category where the bottom-funnel platforms are closed anyway, the argument for building brand and owned reach is stronger than usual, and this is the citation for it.
6. IAB, "Programmatic Advertising: A Close Look at Cannabis". Orientation only.
Published by the Interactive Advertising Bureau's Data Center of Excellence in early 2020, free PDF. Frames cannabis alongside alcohol, tobacco and prescription medicine as a restricted category, and covers programmatic mechanics, brand safety and placement restrictions (no placement near schools or places of worship). INHERITED 2026-08-11 from iab.com.
The problem is the date. 2020 predates the entire current federal picture, every state rule her accounts operate under, and every platform policy revision since. Read it once for the restricted-category framing, which is durable. Ignore every specific. JUDGEMENT.
Not worth the time, paid ads
- WARC and eMarketer. Paid, four to five figures, written for brand-side media planners at scale. Nothing in either addresses a category the platforms refuse. JUDGEMENT.
- Platform-published "playbooks" from Meta and Google. Useful only as policy documents, which her ads knowledge base already covers in section 3. Their strategy content assumes access she does not have. JUDGEMENT.
- Cannabis ad network guides (Mantis, Traffic Roots and equivalents). These are vendor materials describing their own inventory. Fine as a media kit, not as reference. Her ads knowledge base already ranks the channels properly in section 4. JUDGEMENT.
4. Cannabis and hemp ecommerce
The honest verdict first
There is no Baymard for cannabis. There is no Tier A source, no Tier B source, and the Tier C sources describe a different channel from hers. The category's marketing literature is almost entirely agency lead magnets and dispensary-software vendor blogs, and they cite each other, which is why "cannabis marketing best practice" reads the same on twenty sites and traces back to nothing.
What is authoritative in this category is regulatory, not marketing. The regulators, Congressional Research Service, the FDA, the FTC and NAD publish real primary material. Nobody publishes credible research on how to sell this stuff.
The strategic consequence is the most useful finding in this document: acquire the regulatory primary sources, and build the marketing knowledge from her own account data, because nobody else is going to publish it. JUDGEMENT.
A second consequence worth stating: this is a genuine competitive position, not a gap. Any agency competing for these accounts is reading the same lead magnets. A knowledge base built from real send results and a real warning-letter corpus is something none of them will have.
Ranked
1. Congressional Research Service. Free, non-partisan, the best non-vendor source in the category.
Already her source for the federal hemp redefinition (IN12620, LSB11381, IF13136 via congress.gov). Written for legislators by non-partisan analysts, no client, no product, no advocacy position. When a law firm blog and a CRS report disagree, CRS is right. Free. VERIFIED as already in use, cannabis-ads-strategist knowledge base section 12.
What it improves. Nothing new to acquire. Named here so it stays ranked first when a newer, glossier source appears.
2. State regulator primary pages. Texas DSHS, California DCC.
Free, primary, already in use. The only citable authority on what a given state actually requires. Everything else in the category is someone's summary of these pages. VERIFIED as already in use.
3. CANNRA, the Cannabis Regulators Association. Free, and thinner than it looks.
What it is. A national association of state cannabis regulators. Publishes policy documents and fact sheets at cann-ra.org/policies-and-factsheets. Live inventory read 2026-08-11:
- Overview of DOJ Final Order on Marijuana Rescheduling (April 2026)
- "Full Spectrum" CBD Factsheet (2025)
- President's December 18th Executive Order Briefing (December 2025)
- Schedule III Rescheduling Implications Briefing (December 2025)
- 2025 Congressional Appropriations Language on Hemp Briefing (November 2025)
- Best Practices for Regulating Cannabinoids for Safety (2024)
- Best Practice Recommendations for Proficiency Testing Regulations (2024)
- Cannabinoid Hemp Overview, brief and long versions (2023)
- Cannabinoid Hemp Regulatory Challenges, brief and long versions (2023)
- Three members-only documents: environmental regulators guidance, cultivation odor management, medical cannabis programs
VERIFIED 2026-08-11, cann-ra.org/policies-and-factsheets read directly. RECHECK-BY 2026-11-11.
The finding that matters, and it is a negative one. There is no CANNRA fact sheet on advertising, marketing, packaging, labeling or age verification. Secondary write-ups imply CANNRA covers advertising best practice. On the published inventory as of 2026-08-11, it does not. VERIFIED 2026-08-11 by reading the full published list.
What it is good for. Regulatory context and the direction of travel, written by the people who write the rules. Free. Read the "Full Spectrum" CBD factsheet and the cannabinoid hemp overviews once.
What it is not good for. Anything about how to market.
4. NCSL State Cannabis Policy Enactment Database, and the Marijuana Policy Project tracker.
NCSL maintains a database of enacted state cannabis legislation across 50 states, DC and the territories, with published criteria for what counts as a comprehensive medical marijuana law. Funded by state legislatures, neutral by design. Free. MPP monitors and analyses cannabis bills in all 50 states, DC and the territories. Free. Both INHERITED 2026-08-11 from ncsl.org and mpp.org search summaries. RECHECK-BY 2026-11-11.
The difference, and it matters. NCSL is neutral. MPP is an advocacy organisation whose stated purpose is changing marijuana laws. MPP's tracking is genuinely thorough and often faster, but its framing has a direction. Use MPP for coverage and NCSL for neutrality, and never cite MPP to a client as though it were a neutral body. JUDGEMENT.
What it improves. A cross-state view when a store's shipping map spans more than Texas and California, which is the case the ads knowledge base flags as a copy risk.
5. Rice University Baker Institute, hemp legal status mapping. Worth one look.
An academic mapping of hemp product legal status across US states, published by the Baker Institute for Public Policy. Academic and non-commercial, which makes it the only non-vendor, non-regulator source found in this category. The page returned HTTP 403 to an automated fetch on 2026-08-11, so nothing about its currency, coverage or methodology is confirmed here. INHERITED 2026-08-11 from a search result title only. Open it manually before relying on it.
6. Headset industry reports. Free, real data, wrong channel.
What it is. Free market reports built from dispensary point-of-sale integrations: category analyses (beverages, pre-rolls, vape, edibles, concentrates, topicals), state deep dives, pricing and seasonality, demographics, retention. Latest visible report "The Retention Gap" (May 2026). VERIFIED 2026-08-11 that the reports are published and appear free at headset.io/industry-reports. No methodology is disclosed on that page: no retailer count, no state coverage, no sample definition. VERIFIED 2026-08-11 by reading the page.
The disqualifying caveat, stated plainly. Headset's data comes from licensed dispensary POS systems. Her accounts are hemp-derived DTC ecommerce shipping to consumers. Different channel, different legal regime, different buyer, different basket. A Headset finding about what sells in a Michigan dispensary says nothing reliable about who buys from tghhouston.co.
What it is genuinely good for. Category-level product and format trends (which formats are growing, seasonality shape). Treat as directional context, never as an audience insight, and never quote a Headset number as though it described her list. JUDGEMENT.
7. Self-regulatory codes: Hemp Beverage Alliance, US Hemp Roundtable, ATACH.
Free, thin, and mostly principle statements rather than operational guidance. The Hemp Beverage Alliance publishes marketing principles including a prohibition on targeting or promoting to people below legal drinking age across all marketing including social. The US Hemp Roundtable has partnered with an identity verification vendor to bring age gating to members. ATACH works on distinguishing intoxicating from non-intoxicating and synthetic from natural products. All INHERITED 2026-08-11 from secondary sources and association pages surfaced in search; none opened directly. RECHECK-BY 2026-11-11.
Why they are on the list at all. These are what a client or a client's counsel will point at when they want a defensible standard. Knowing the codes exist and what they say is worth an hour, mostly so a recommendation can be framed as meeting an industry code rather than as her opinion.
Not worth the time, cannabis
- MJBiz Factbook, BDSA, Brightfield Group, New Frontier Data. Paid market intelligence, four figures and up. They sell market sizing, TAM, and category forecasts. She does not need to know how big the market is. She needs to know what sentence to write in a subject line. Wrong product for the problem. JUDGEMENT. Brightfield publishes one free report, which is a lead magnet. INHERITED 2026-08-11.
- NCIA "Social Media Guidelines for Cannabis Marketers". Free PDF, but dated 2021. It predates every current platform policy, the 2025 federal hemp redefinition, Texas's 2026 reclassification and California AB 8. Actively misleading now. INHERITED 2026-08-11 from the hosted PDF date.
- Agency and software-vendor guides (MoonSauce, Cannabrands, Cova, Kepler Leaf, Hybrid Marketing, Just Digital Gurus, Flowhub, and the rest). Tier D. Recycled, unsourced, mutually citing, and collectively responsible for the circular "best practice" in this category. Two partial exceptions worth keeping as pointers only, never as citations: Highfloor and cannabisregulations.ai have both been useful for surfacing primary sources quickly, and both already appear in her ads knowledge base source list in that role. JUDGEMENT.
- Law firm blogs (Vicente, Harris Sliwoski, Perkins Coie, Buchanan Ingersoll). Genuinely the fastest read when a rule changes, and already in her sources. But they are marketing for legal services and they are written to be quotable, not to be complete. Use them to find the statute, then cite the statute. JUDGEMENT.
5. Document types worth building beyond the Rise five
She asked what document types would actually help. The Rise set is five documents: Brand Voice Guide, Customer Avatar, Compliance Guidelines, Offer Matrix, Product Catalog. That set answers who the brand is, who buys, what is forbidden, what the deal is, what is sold.
It does not answer what to say, whether it has already been said, and whether it worked. That gap is precisely where her stated problem lives, because voice and angle are exactly what the Rise five underspecify.
Existing coverage, so nothing below duplicates it: a Reddit signal feed (customer language in the wild), a knowledge curator (freshness discipline on the knowledge bases), and two agent knowledge bases (Klaviyo strategist, ads strategist).
Per-brand, extending the Rise five
6. Banned and Approved Phrase List. The highest-value single document to build. A flat lookup table: exact phrase, verdict (banned / conditional / approved), the reason, and the citation. Sources: the FDA warning letter corpus, the FTC substantiation standard, ESP acceptable-use policy, platform ad policy, and every line her own QA has already rejected.
Why this and not another guideline document: her problem statement is that copy "gets caught in QA instead of being right at the source". A principle ("no therapeutic claims") does not stop a draft, because the writer does not think "finally rest" is a therapeutic claim. A phrase list does stop it, because it is a string match. This moves compliance from a review step to a write-time constraint, which is the only structural fix to the problem she described. JUDGEMENT.
7. Angle Bank. Not personas. Angles. One entry per angle with: the angle in one line, the segment it is aimed at, the proof or product truth it rests on, every send it has been used in with dates, what it earned (RPR), and a burnout flag.
The Rise Customer Avatar tells an agent who it is writing to. Nothing in the Rise set tells it what to say or whether that thing has already been said three times this quarter. This is the document that fixes "misses on angle". It also becomes structurally necessary on TGH, where the standing ruling is no discount code at all: a send that previously ran on 40% off now has to carry a real angle, and her own note already says a send with nothing left to say needs rebuilding rather than shipping weak. An angle bank is what makes that rebuild fast instead of a blank page.
8. Objection and Question Inventory, sourced from real customer contact. The Rise Customer Avatar has an objections section, but its objections are client-confirmed assertions, which is someone's memory of what customers say. Hers can be built from actual support tickets, product reviews and on-site search terms, which is stronger evidence than any external source on this list and is unavailable to any competitor. Complements the Reddit feed rather than duplicating it: Reddit gives category language from strangers, this gives objection language from buyers.
9. Competitor Send Archive with verdicts. Dated captures of what competitors actually send, each annotated with what is good, what is copyable, and what would fail QA here and why. The "what would fail here" column is the part that makes it a training document rather than a swipe file.
10. Send and Test Ledger. One row per send: date, angle used, format (text or design), segment, RPR, and a one-line lesson. This is the document that makes the whole knowledge base self-improving from real results rather than from intent, which her standing rules already require. It is also the only thing that will ever settle the text-versus-design question on TGH properly, since the current 30.5% versus 17.1% figure is flagged in her own knowledge base as produced at a bad API revision and needing re-derivation.
Shared playbooks, extending the five in the Rise set
11. CRO Playbook for Age-Gated Ecommerce. This document does not exist anywhere externally. The interaction between an age gate and a conversion funnel is not covered by Baymard, NN/g or any cannabis source found here. Everything needed to write it exists: NN/g checkout volumes for the funnel, the state age requirements already in her knowledge bases, and her own site observations. Genuinely novel, and directly useful to the CRO work routed to Sehrish Iqbal.
12. Measurement Without Attribution. Her ads knowledge base section 5 is already a draft of this. Promote it to a standalone playbook: triangulate three views, post-purchase survey, UTM discipline, when incrementality is and is not readable at a given spend. The last part is the one nobody else writes down, and it is the part that stops a client being promised a test that cannot produce a signal.
13. Restricted-Category Channel Playbook. Ads knowledge base sections 3 and 4 promoted: which platforms refuse this category, which certified paths exist and what they cost, and where the reachable inventory actually is, ranked. Durable across brands, and it is the document that prevents an agent proposing a Meta campaign that cannot run.
14. Pre-Flight QA Rubric, as a document. The checklist a specialist agent runs against its own output before returning it. This already exists as a working practice on Salvo, in the Karissa rules. The TFC equivalent should be written down rather than carried in her head, and it is the natural place to bolt the phrase list (item 6) on as an automated gate.
Do not build
- A market-sizing or category-trends document. Duplicates paid vendors, adds nothing to a send.
- Another persona document. The Rise Customer Avatar shape is sufficient; the gap is angles, not personas.
- A general "cannabis marketing best practices" document. That is the exact Tier D artifact this file argues against, and writing our own would not make it evidence.
6. Recommendation
Acquire now, free, this week
Ranked by value per hour spent:
- FDA cannabis-derived warning letters. Pull the corpus, extract the cited claim language. This is the input to the phrase list and it also closes item 12 in the knowledge base's unsubstantiated-facts list.
- FTC Health Products Compliance Guidance, plus the Endorsement Guides. Read once, cite forever. Gives the claims rule a real citation.
- NAD decision summaries, cannabinoid cases. Free tier is enough.
- CANNRA free fact sheets. One hour, regulatory context, and the negative finding above is now recorded so nobody hunts for advertising guidance there again.
- HTTP Archive Web Almanac 2025, ecommerce and performance chapters. One hour, free, settles platform and speed arguments with real data.
Acquire with money, in order
- NN/g ecommerce volumes, four of them, roughly $400 once. Suggested: Shopping Carts and Checkout, Product Pages, Trust and Credibility, Transactional Email and Notifications. No renewal, no subscription, and Trust and Credibility is unusually relevant to a category where the buyer's first question is whether the store is legitimate.
- GoodUI Solo, roughly $60 to $72/mo, only if she commits to actually running tests. The effect sizes are the product and they are behind the paywall. Confirm live pricing first, since it was not visible on the public site.
- Baymard Expansion, $4,200/yr. Only when CRO outweighs email in her workload. Foundation at $2,400 is the wrong tier because it excludes cart and checkout, which is the half that matters for a hemp DTC store. Until then the free tier plus NN/g covers it.
Do not buy
MJBiz Factbook, BDSA, Brightfield, New Frontier, WARC, eMarketer, CXL. Between them they answer questions nobody at TFC is asking, at a price that would buy every NN/g volume several times over.
Build, in this order
- Banned and Approved Phrase List (per brand, shared skeleton). Highest leverage, because it is the only item that attacks the stated problem structurally.
- Angle Bank (per brand). Second highest, and TGH needs it immediately given the no-code ruling.
- Send and Test Ledger (per brand). Cheap to start, compounds, and it is the only path to re-deriving the text-versus-design question honestly.
- Measurement Without Attribution and Restricted-Category Channel Playbook (shared). Mostly promotion of existing knowledge base sections, so low cost.
- CRO Playbook for Age-Gated Ecommerce (shared). Wait until the NN/g volumes are in hand.
The one-line version
Buy about $400 of Nielsen Norman Group, spend one day pulling free primary regulatory material, and build the phrase list and the angle bank. The rest of the money on this page buys documents that answer questions she is not being asked.
7. Recheck schedule
| What | Cadence | Next due |
|---|---|---|
| Baymard, NN/g, GoodUI pricing and free-tier boundaries | 90 days | 2026-11-11 |
| FDA warning letters for cannabis-derived products (new letters) | 30 days | 2026-09-10 |
| NAD cannabinoid decisions | 30 days | 2026-09-10 |
| FTC Negative Option Rule status (currently INHERITED) | 30 days | 2026-09-10 |
| CANNRA published fact sheet inventory | 90 days | 2026-11-11 |
| Web Almanac (2026 edition) | 90 days | 2026-11-11 |
| Meridian and GeoLift (only if spend ever reaches a readable level) | dormant | when spend changes |
| Tier assessments and rankings in this file | never, they are judgements | revise on new evidence |
8. Open items this file creates
- Baymard free statistics pages were not opened directly. The 70.22% abandonment figure, the 55% to 84.27% range, the 39% / 17% reason splits and the 23.48 form elements benchmark are all INHERITED from a search summary. One page load promotes all of them.
- The FDA warning letter page was not opened directly. Everything in section 3 item 1 rests on the search result set. Open it before building the phrase list from it.
- The FTC guidance page itself was not opened, only its press release and four law firm analyses, which agree. Cheap to close.
- GoodUI pricing is unconfirmed. The public site gates it. Numbers quoted are secondary.
- The Baker Institute hemp mapping returned 403 to an automated fetch. Needs a manual look before it can be ranked properly.
- NN/g report publication dates are not shown, only edition numbers. Worth asking or checking before a finding from an old edition becomes a rule.
- Headset publishes no methodology on its reports page. If any Headset figure is ever used, the methodology has to be obtained first.
9. Sources
All accessed 2026-08-11. Marked (read) where the page itself was opened, (search) where the finding comes from a search result summary only.
CRO
- baymard.com/pricing (read); baymard.com (read, via pricing page); baymard.com/lists/cart-abandonment-rate (search)
- nngroup.com/reports/topic/e-commerce (read); nngroup.com/reports/ecommerce-ux-product-pages-including-reviews (read); nngroup.com/reports/free (search)
- goodui.org (read); goodui.org tests and patterns (read)
- almanac.httparchive.org/en/2025/ecommerce and /performance (search)
- analytics-toolkit.com and abtestingstats.com (search)
- measuringu.com and uxpajournal.org SUPR-Q paper (search)
- growth.design (search)
Paid ads and claims law
- ftc.gov Health Products Compliance Guidance and the December 2022 press release (search)
- Morrison Foerster, Cooley, Wilson Sonsini, Venable analyses of the FTC guidance (search)
- fda.gov Warning Letters for Cannabis-Derived Products (search)
- bbbprograms.org/media/newsroom/decisions (read); bbbprograms.org CannaPure CBD referral (search)
- blog.google Meridian announcement; pypi.org/project/google-meridian (search)
- github.com/facebookincubator/GeoLift (search); research.getrecast.com geo-experiment simulation study (search)
- ipa.co.uk, The Long and the Short of It presentation (search)
- iab.com, Programmatic Advertising: A Close Look at Cannabis, 2020 (search)
Cannabis and hemp
- cann-ra.org/policies-and-factsheets (read)
- ncsl.org State Cannabis Policy Enactment Database (search); mpp.org/policy (search)
- bakerinstitute.org hemp legal status mapping (403, not read)
- headset.io/industry-reports (read)
- hempbeveragealliance.org/principles (search); US Hemp Roundtable and ATACH via secondary reporting (search)
- thecannabisindustry.org NCIA Social Media Guidelines for Cannabis Marketers, 2021 (search)
- cannabisregulations.ai NAD 2025 analysis (search, secondary)
Internal
C:\Users\marit\dev\klaviyo-cannabis-strategist\knowledge.mdC:\Users\marit\dev\cannabis-ads-strategist\knowledge.mdC:\Users\marit\Downloads\Email Marketing Trainer\(Rise template set, 10 files)