Keyword research with Semalt AI: why volume alone stopped mattering in 2026

The problem with volume-only research

For decades keyword selection had a simple recipe: filter by high volume, sort by low difficulty, pick the ones that look reasonable. That approach produces two failure modes we see every month:

• Ranking without selling. One ecommerce hit top 3 for «cheap furniture» (2,900 monthly searches). Got 1,100 extra monthly visits. Sold four tables. The term attracted price-comparison searchers, not buyers.
• Ranking never possible. A fintech startup targeted «online bank account» (14,800 monthly searches). Four competitors are banks with DA 88+. It never passed position 34.

Volume is vanity. What matters is the combination of right intent, expected conversion value, competition you can actually beat and a SERP where organic clicks still exist. Semalt AI evaluates all four simultaneously.

The 4 dimensions Semalt AI evaluates

1. Search intent classification with a semantic LLM

Semalt AI's classifier (proprietary fine-tune on Claude Sonnet 4.5) labels each keyword into 5 buckets:

• Pure informational — «what is seo». Big traffic, zero direct conversion, valuable only for brand or linkbait.
• Pre-purchase informational — «how to choose an seo agency». High value: the user is researching and a solid guide converts.
• Comparative — «semrush vs ahrefs pricing». Very high conversion if your product is in the comparison.
• Transactional — «hire seo agency santiago». The crown.
• Navigational — «santiago seo login». Only useful for your own brand.

This mapping runs in batch. Feed it 8,000 keywords and 20 seconds later you get the distribution. Discarding the 60% that turns out to be pure informational frees hours of misdirected editorial work in most clients.

2. Expected value per conversion (EVC)

If you connected the CRM inside Semalt Analytics, the engine knows your average ticket, conversion rate per intent bucket, and sales cycle. It calculates expected revenue per organic click for each keyword. Keyword A can have 1,200 monthly searches and an EVC of USD 4, while keyword B has 180 monthly searches and EVC of USD 90 — B is almost 4x more valuable despite lower volume.

3. Real difficulty by sector

Traditional KD (Keyword Difficulty) measures top-10 domain authority. Semalt AI adds three factors that change everything:

• Freshness of top-ranking content (a 2019 page not updated is vulnerable).
• Coherence between the keyword and the ranking titles (if the top 10 does not match intent, there is a hole).
• Local vs global site presence (for local queries, if the top 10 is dominated by Spanish or Mexican sites and you are a well-optimised local site, you have an edge).

4. SERP features and organic space

Google in 2026 shows AI Overviews on 41% of informational queries. The organic click is pushed down, sometimes below the mobile fold. Semalt AI marks for each keyword which SERP features show up (AI Overview, People Also Ask, video, map, shopping) and estimates realistic organic CTR. A keyword with AI Overview + video + map may have position-1 CTR of just 8% instead of the traditional 28%. Semalt weighs that in.

Step-by-step: from 8,000 keywords to 30 actionable clusters

This is the flow we run for every new client. It takes between 90 minutes and one afternoon.

Step 1: Initial seed (10 min)

In Semalt's Keyword Research tab, enter 3-8 seed keywords describing your business. For a local dentist: «dentist santiago», «dental implant chile», «orthodontics», «pediatric dentist». The engine returns 3,000 to 12,000 related terms including long-tail variants, questions and comparatives.

Step 2: Intent filter (2 min)

Apply «Bucket = Comparative OR Transactional OR Pre-purchase informational». Discard navigational (competitor brand) and pure informational (unless you are in a content-marketing phase).

Step 3: EVC filter (1 min)

Sort by EVC descending. Review the top 100. Usually a mix of comparatives and transactional-specific-by-neighbourhood or sub-service.

Step 4: Winnability (5 min)

Apply «Adjusted difficulty < your estimated authority + 15». Semalt calculates your authority automatically. This filter kills impossible fights but keeps ambitious ones that a small link-building push resolves.

Step 5: Realistic SERP (10 min)

Sort by «estimated organic CTR × volume» and discard keywords with AI Overview + shopping + map on top (those hand out almost no clicks). Remaining: 80-200 keywords.

Step 6: Clustering and mapping (25 min)

The «Cluster mode» groups keywords sharing the same ranking URL (if two keywords have the same top 3, they should be attacked with a single page). Output: 20-40 clusters, each with a primary keyword and 3-10 secondaries.

Step 7: Final prioritisation (10 min)

Sort clusters by total EVC and split into «first wave» (30-70% of EVC, 8-15 clusters), «second wave» and «explicit long-tail». The first wave is your 90-day content roadmap.

Optimising for local SERPs

Google.cl is not simply Google.com with a different domain. Local SERPs have their own idiosyncrasies:

• Marketplace dominance in retail transactional queries. Semalt AI flags red when marketplaces occupy more than 3 top-10 slots — it tells you you are not competing via SEO but by getting your product into those marketplaces alongside your own store.
• Local slang: real classifiers understand regional Spanish. The engine suggests variants with more local intent.
• Aggressive local pack for queries with district names. Semalt marks which of your keywords are local-pack-eligible and reserves those for Google Business Profile work.
• Low competition in regional Spanish vs Spain: content written in genuine Chilean register (not translated, not copied from Spanish sites) has natural advantage over translations.

Hidden keywords: SGE, People Also Ask and Discussions

The search engine no longer shows only 10 links. It shows AI Overviews with citations, People Also Ask with related questions and Discussions with Reddit / forums. Each of those «layers» is a traffic opportunity classic tools do not report. Semalt AI extracts:

• PAA questions for each keyword in your shortlist (typically 4-12 per keyword). Each is a FAQ or standalone-article opportunity.
• AI Overview citations: which sites Google cites in the summary. Very high value: if you are cited, you gain traffic and authority; if a competitor is cited, you know exactly who to beat.
• Reddit/Quora/forum threads Google promotes in Discussions. Each is an opportunity for your own positioning (expert posting) or research (what pain users express).

Case: boutique real-estate agency, +214% leads in 5 months

Boutique real-estate agency in an upscale district, focus on apartments in the $250k-$800k range. They had 43 «target» keywords inherited from a previous consultancy. Aggregate volume looked great (24,000 monthly searches). Leads generated in 12 months: 61. Cost per lead: expensive.

We applied the Semalt AI method. Research result: 16 prioritised clusters, none in the original set.

• «new apartment vitacura ready to move in» (high EVC, medium competition, 8 PAA questions)
• «apartment with rental guarantee santiago» (comparative, very high EVC)
• «ds01 subsidy new apartment santiago» (niche, medium EVC but zero competition)
• … +13 similar clusters

They shipped 16 new pages plus 4 refreshes over 2 months. Result at 5 months: 192 qualified leads (3.1x in volume, 4.9x in intent quality). Cost per lead dropped 68%. The 43 original terms? The team stopped working them. Sessions from those pages: down 12%. Attributable revenue: up 214%.

More in this series: Semalt Analytics: the SEO data GA4 hides, Semalt AutoSEO: optimisation that runs 24/7, Competitor analysis without leaving a trace.

FAQ

Do I need to connect my CRM?

Not mandatory but triples the value. Without CRM, Semalt AI uses industry benchmarks to estimate EVC. With CRM, EVC is real and personalised. Connection takes 10 minutes via webhook.

How many keywords does the standard plan process?

500,000 keywords per month with full AI classification. Enough for agencies of 20-40 accounts or major brands. Above that, enterprise gives unlimited processing and queue priority.

Does it work for multiple languages?

Yes. The classifier works natively in 27 languages including English, Brazilian Portuguese, French and all regional Spanishes. Especially useful for LATAM agencies comparing the same term across different markets in a single dashboard.

Can I export to CSV or Google Sheets?

Yes. Direct CSV/XLSX export and a native Google Sheets connector that refreshes every 6 hours. For agencies working with Google Data Studio, there is an official Looker Studio connector.

Are volumes updated in real time?

Volumes update monthly for low-frequency keywords and weekly for the top of the sector. Intent and AI classification recalculate every time the SERP shifts significantly (typically 1-3 times per month per keyword).