Most CBD keyword research dies on day one. The marketer pulls “best CBD oil” from Ahrefs, sees 60k volume and KD 89, picks something easier called “best CBD oil for anxiety” at KD 71, and accepts they’re going to write 50 articles to chase it.
That’s how CBD content marketing has worked for five years and it’s why most CBD blog libraries look identical. It also doesn’t work — those 50 articles compete with 5,000 identical articles from older brands and never rank.
The vertical-first approach is different. Three clusters tell you what to write before you ever look at a keyword tool.
The four cannabinoid pillars (and the THC pillar)
A modern CBD brand sells more than CBD. The Farm Bill umbrella covers CBD, CBG, CBN, CBC, hemp-derived delta-9 THC, delta-8 and HHC. Each is its own pillar with its own keyword cluster.
CBD pillar: the saturated one. Head terms are gone — owned by the big DTC players (Charlotte’s Web, Lazarus Naturals, CBDistillery, Joy Organics, Bluebird Botanicals). Realistic targets: mid-tail formats (“water-soluble CBD”, “CBD for [specific condition with proper compliance”), “broad-spectrum vs full-spectrum” comparison content.
CBG pillar: the open one. As of 2026, ~80% of cannabinoid SEO content is still about CBD. CBG (“the mother cannabinoid”) has order-of-magnitude smaller competitor density. Brands that own CBG content clusters now will hold them through 2027–2028 when the broader market catches up.
CBN pillar: the sleep-vertical opportunity. CBN searches are tied tightly to sleep-aid intent, which makes them commercially valuable per query. Three-to-four established competitors total in the cluster as of mid-2026.
CBC pillar: the niche. Smaller search volume than CBN, even thinner competition. Best as a layer on top of an established CBD/CBG foundation.
Hemp-derived THC pillar: the volatile one. Federally legal under the 2018 Farm Bill if ≤0.3% delta-9 by dry weight. ~16 states have banned delta-8 as of 2026; more restrict hemp-derived delta-9. The 2024–2026 Farm Bill renewal could close the loophole entirely. We rate this pillar as “play if you have to, hedge with CBD/CBG underneath.”
The regulatory layer (where most brands won’t compete)
Compliance-language queries are commercially valuable and almost nobody competes for them properly:
- “is CBD legal in [state]” — 50 possible state-pages, all with consistent intent.
- “does CBD show up on a drug test” — high-volume head term, and Yes the answer should be honest.
- “CBD drug interactions” — buyer with a doctor’s appointment tomorrow.
- “CBD dosage [condition]” — without making medical claims.
- “FDA CBD warning letters [year]” — the brand that maintains a current list gets cited everywhere.
- “[product] vs [competitor product]” — comparative content that AI engines lift verbatim.
- “best CBD brands certified by [third-party verifier]” — owns your COA and review-platform proof.
Most CBD content libraries skip these because they don’t show up in keyword tools the same way as product-format keywords (“CBD gummies for sleep”). They show up in People Also Ask, in Quora, in Reddit, in ChatGPT prompt logs (if you have them).
We mine these from three sources: Google’s PAA on the head terms, the search console internal-search log for sites that have it, and prompt-tracking tools like Profound and Otterly.AI that surface what people are actually asking AI engines.
The cluster blueprint
Once the pillar is picked and the regulatory layer is mapped, the cluster has a predictable shape:
- 1 pillar page (3,000–5,000 words) — the definitive overview of the cannabinoid.
- 6–10 buyer-intent pages — “X for [condition]” with proper FDA-disclaimer compliance, “X dosage”, “X side effects”.
- 15–30 long-tail FAQ pages — single-question pages each addressing one specific compliance or product question. ≤30-word direct answer + 200–400 words of evidence-backed depth.
- 5–10 comparison pages — “X vs Y”, “X vs CBD”, “X vs prescription”.
- 3–5 state-restriction pages — for the cannabinoid where state restrictions apply (delta-8, hemp-derived delta-9).
Per cluster total: 30–80 supporting URLs, deployed across 6–18 months depending on tier and content-team capacity.
How AEO changes keyword research
AEO citations follow a different logic than rankings. ChatGPT picks one source per concept; Perplexity picks 3–7. Google AI Overviews picks 2–5. The brand that owns the cleanest factual paragraph on a concept gets cited disproportionately, even if its rank in classic SERP is page 2 or 3.
That changes the workflow. Traditional keyword research starts from search volume. AEO-first keyword research starts from buyer questions, then checks if those questions trigger AI Overviews, then counts how many competitors are currently being cited.
Specifically:
- Pull the top 100 buyer questions from PAA, Reddit, Quora and prompt-tracking tools.
- For each question: does ChatGPT answer it? Does Perplexity? Does Google show AIO?
- For the questions that fire AI search: how many competitor brands does the AI cite? (Often 1–3 — that’s the citation slots you need to take.)
- For the questions that don’t fire AI search: are they likely to in 12 months? (Compliance and dosage queries are increasingly AIO-eligible as of 2026.)
- Prioritize questions that fire AI search AND have ≤3 cited competitors AND have meaningful commercial intent.
That priority order produces a 60–120 question target list that becomes the FAQ-page production calendar for the next 12 months.
The keyword research deliverable we ship to clients
Foundation tier: a 200–300 keyword list with cluster mapping, search volume, KD, AIO/AEO eligibility per query, and target page mapping.
Growth tier: same plus weekly AIO-presence tracking on 30 priority queries, plus Reddit/Quora/Profound mining for emerging buyer-intent prompts.
Scale tier: same plus original prompt-research on 60+ queries, plus quarterly research-piece publication that becomes the cited primary source for AI engines on the cluster.
Want a sample CBD keyword research output? Email Marcus and ask for the redacted CBG-pillar export from a recent client. Plano-Texas based, no sales pitch attached.