What this is
One row per canonical search term. 26,250 US English terms about psoriasis, psoriatic arthritis and psoriasiform disease, each carrying a demand band, a 48-month shape, a trend direction and five facets.
What this is not
- Volumes are bands, not figures. Google Ads reports the midpoint of a bucket on a fixed geometric ladder, so its own numbers are estimates. We publish the band those estimates fall in, deliberately coarser than the source ladder.
- The demand index is within-term only. Each series is scaled 0-100 against that term's own peak. Comparable across months for one term, never between terms - use the volume bands to compare terms.
- Rows are canonical terms, not raw phrasings. Google groups close
variants onto one bucket;
variant_counttells you how many phrasings collapsed into each row. Do not sum volume across rows without accounting for it. - stable and insufficient_data are different claims. One means we measured and it did not move; the other means we could not measure.
- This is education-and-discovery demand. 74% of terms sit in treatment-discovery or condition-understanding and only 4% name a brand. It is not a brand-share dataset.
Columns
| Column | Type | Meaning |
|---|---|---|
| keyword | STRING | The canonical search term |
| volume_lower / volume_upper | INT64 | Monthly-searches band, lower inclusive, upper exclusive |
| volume_rank | INT64 | Rank by volume, 1 = highest. Filter on this for top-N |
| demand_index_series | STRING | 48 comma-separated integers 0-100, 2022-08 to 2026-07 |
| data_through_month | STRING | Vintage of the series (2026-07) |
| trend_label | STRING | rising_fast 3.0% · rising 2.2% · stable 53.1% · declining 8.6% · declining_fast 2.0% · volatile 1.6% · insufficient_data 29.6% |
| trend_confidence | STRING | high · medium · low · none |
| growth_rank | INT64 | Rank by change among directional terms |
| journey_stage | STRING | treatment_discovery 38.7% · condition_understanding 35.2% · ongoing_management 8.9% · treatment_evaluation 6.3% · differential_seeking 4.8% · symptom_noticing 3.5% · treatment_access 2.7% |
| intent_type | STRING | informational 49.1% · product_discovery 28.1% · procedural_clinical 9.0% · comparison 5.6% · brand_navigational 4.4% · community_social 3.7% |
| brand_signal | STRING | unbranded_disease 82.1% · unbranded_product_category 13.3% · branded_consumer 2.5% · branded_rx 1.5% · unbranded_drug_class 0.7% |
| clinical_focus | STRING | general 66.4% · psoriatic_arthritis 13.6% · scalp 10.5% · plaque · nails · face · hands_feet · guttate · trunk_limbs · genital · inverse · pustular · erythrodermic |
| buyer_relevance | STRING | Pipe-separated, one or more of: health_education, content_seo, provider_hco, consumer_health, retail_pharmacy, general_market, pharma_competitive, payer_access, pharma_brand, investor_signal |
| drug_mentioned / drug_class | STRING | Named therapy (27 in the vocabulary, generics resolved to brand) and its class |
| rank_in_journey_stage / _clinical_focus / _brand_signal / rank_for_drug | INT64 | Volume rank within each facet - filter for per-segment top-N |
| variant_count | INT64 | Phrasings that collapsed onto this canonical term |
| word_count / is_long_tail / is_question | INT64 / BOOL | Query shape |
Method
Volume bands. Every second point of Google's published ladder is a band edge, so each band spans two ladder points and no single reported value is recoverable. Edges: 10, 30, 50, 90, 140, 210, 320, 480, 720, 1000, 1600, 2400, 3600, 5400, 8100, 12100, 18100, 27100, 40500, 60500, 90500, 135000, 201000, 301000, 450000, 673000, 1000000, 1500000, 2240000, 3350000.
Trend. Median of the last 12 months against the median of the previous 12. Medians so a single spike cannot move them, a full year each side so seasonality cancels. The threshold varies by volume because one ladder step is +100% at the bottom and ~22% at the top; a term must move further than the source could express by rounding before it is called a direction. That bar is applied per term during the build; the label is what ships.
Classification. journey_stage and intent_type are LLM annotations (structured output, temperature 0); median confidence 0.80, 2.7% below 0.65. brand_signal, clinical_focus and drug_mentioned are deterministic matches against controlled vocabularies. buyer_relevance is derived by rules.
Example queries
?volume_rank=<=200 top 200 terms by demand ?clinical_focus=scalp&rank_in_clinical_focus=<=50 top 50 scalp terms ?trend_label=rising_fast&trend_confidence=high confident risers ?journey_stage=treatment_access cost and coverage demand ?drug_mentioned=skyrizi one therapy's full term set ?brand_signal=unbranded_disease&is_question=true unmet education demand
Coverage
United States, English. Series 2022-08 to 2026-07 (48 months). 26,250 terms measured from a 108,953-term discovery pool; 8,599 returned no data, 2,727 were close-variant duplicates and 874 sat at zero volume.
Known limitations
- 84% of terms fall in the 10-30 band. Google's reporting floor is 10, so the long tail is not further resolvable. The value is coverage and faceting, not volume precision.
- 30% of terms are insufficient_data for trend - too few active months in the compared windows to make a claim.
- clinical_focus is general for 66%; most terms name no specific aspect.
- Brand tagging is list-based. A therapy outside the 27-drug vocabulary is not tagged.
- Reported search volume is not automatically ground truth - bot and AI traffic can inflate it. 2,034 terms carry a non-seasonal spike of 5x or more above their median; these are retained, not removed.
- US English only, and a periodic re-cut rather than a live feed.
Refresh
Periodic re-cut, not a live feed. data_through_month carries the vintage so your pipeline can detect staleness without reading this page.