Keyword research with AI: Semalt scores Arabic and English terms by expected Riyal, not by MSV

Why search volume alone is a trap in KSA

Every Riyadh SEO knows the story. A junior brings a keyword list to the standup: “افضل موبايل 2026” at 42,000 monthly searches, KD 3 CPC, low difficulty, everyone excited. Six months later the URL ranks position 5, gets 4,100 clicks a month, and generates precisely 1,800 SAR of revenue — because the query is dominated by teenagers doing homework research on youtube.com.sa results, not buyers. Meanwhile the head of e-commerce quietly tracks a 90-searches-per-month query — “افضل موبايل للتصوير تحت 2500 ريال” — that generates 78,000 SAR/month because every single searcher is a buyer with a budget.

Monthly Search Volume (MSV) is the wrong ranking function for Saudi SEO. Google Keyword Planner numbers are also aggressive round-offs and pull from google.com not google.com.sa. The right ranking function is expected revenue in SAR, computed from realistic CTR-by-position curves, realistic CVR-by-intent, and the buyer's real price band. That is what the Semalt Keyword AI actually returns.

How the Semalt AI model works

The engine is a stack of three models rather than a single one, deliberately — each layer catches a failure mode the previous layer cannot see:

1. A demand model that predicts real monthly search volume on google.com.sa specifically. Trained on a rolling window of Search Console data from 39 anonymised Saudi sites plus a raw query corpus.
2. An intent classifier that assigns each query one of eight intents: informational, commercial-comparison, commercial-buy, navigational, local-service, governmental, Hajj/Umrah, and Vision 2030.
3. A revenue model that multiplies volume × expected CTR-by-position × expected CVR-by-intent × expected order value in SAR (learned from your own connected shop or an industry median if not connected).

The output is a single SAR/month number per keyword, plus a confidence band. You sort the list by that number, and the top 30 rows are what your content plan for the next quarter should look like.

Najdi, MSA and Hijazi dialect classification

Saudi Arabic is not one thing. A Riyadh buyer searches differently from a Jeddah buyer, and both differ from a Modern Standard Arabic (MSA) query written by a news-desk sub-editor. Semalt classifies every query into three dialect buckets before scoring:

The MSA bucket, by contrast, is where news queries and academic queries sit — important for publishers, mostly noise for retailers. Semalt lets you weight the three buckets by your ICP. A Riyadh Salla merchant serving central KSA will typically set Najdi to 1.0, Hijazi to 0.4 and MSA to 0.1. A national news publisher will do the opposite.

Buyer intent, governmental intent and Vision 2030 lists

Semalt's governmental intent classifier fires whenever a query contains one of ~1,400 signals: “وزارة”, “هيئة”, “my.gov.sa”, “absher”, “nafath”, ministry name, initiative name. A retailer usually wants to exclude these — they are not going to convert. A ministry contractor or a Vision 2030 supplier wants to include only these. One checkbox in the AI filter and the list rebalances in 300 ms.

The Vision 2030 list is curated separately and updated monthly — NEOM, THE LINE, Trojena, Qiddiya, Diriyah Gate, Red Sea Global, ROSHN, Green Riyadh, and every officially-named initiative. Any query containing one of those signals inherits a higher CVR prior because searchers are typically B2G decision-makers.

Hajj, Umrah, Ramadan and White Friday seasonality

Generic keyword tools smooth all seasonality into a 12-month rolling average. That is fine in Sweden. It is a disaster in Saudi Arabia, where a query like “افضل شنطة عمرة نساء” sits at zero for eight months and produces 84 % of its annual revenue in the 45 days before Ramadan.

Case study: a Dammam industrial parts seller

A Dammam-based industrial parts distributor — ~140,000 SKUs sold to Aramco contractors, SABIC vendors and Jubail refinery service providers — had been running a Semrush export for two years. Content team publishing three articles per month, ranked by MSV. Twelve months of that programme had produced 1.8 % organic revenue growth — below the general market.

We loaded the same seed list into Semalt Keyword AI, connected the client's Odoo instance so the model could learn real CVRs by category, and re-ranked. The top 30 keywords by SAR/month:

• 16 of the 30 had MSV under 200/month — every single one would have been ignored by the old workflow.
• Only 4 of the 30 overlapped with the previous 12 months' output.
• The top item was a Najdi long-tail with 60 monthly searches and 41,000 SAR of predicted monthly revenue.
• Two Vision 2030 supplier queries (mega-project procurement) came in top 10.

The team switched the calendar. In the following 90 days organic pipeline was up 46 %. The traffic count barely moved. The revenue moved.

Workflow: from brief to ranked list in 25 minutes

Semalt Keyword AI vs. Ahrefs vs. Semrush

More in this series: Semalt Analytics: what GA4 hides from you, Semalt AutoSEO: on-page fixes without a dev, Semalt: backlink gap, keyword gap, alerts.

FAQ

Where does the raw Arabic query corpus come from?

An aggregated, anonymised panel of 39 Saudi-hosted Search Console properties plus a licensed clickstream feed from three GCC ISPs. Individual queries are de-duplicated and pseudonymised before ingestion, and the panel opts in via the standard Semalt DPA.

Does the dialect classifier handle Egyptian Arabic queries?

Yes — but they are bucketed as MSA-adjacent and down-weighted for KSA projects by default. If your ICP includes expatriate communities in Riyadh or Jeddah, we recommend raising the ar-EG weight to 0.2.

How often is the model retrained?

Weekly, with a full retrain every quarter. The intent classifier is retrained monthly against a labelled set of 3,000 fresh KSA queries reviewed by native Riyadh and Jeddah annotators.

Can we plug it into our existing brief tool?

Yes. Every ranked keyword export includes intent, dialect, seasonality curve, competitor snapshot and predicted SAR in a single JSON blob. Native connectors ship for Notion, Airtable, Google Sheets, ClickUp, Asana, monday.com, Trello and Jira.

Does it integrate with Semalt AutoSEO and Analytics?

Yes. Send a keyword to AutoSEO with one click and it becomes a page brief, then a live optimisation run. The revenue delta is measured by Semalt Analytics and closes the loop.