How We Mapped 89,000 Search Terms Using AI (And What We Found)

Most SEO agencies map 200 keywords and call it research. We mapped 89,000. And what we found changed how we think about search entirely.

This isn't a story about software. It's a story about what happens when you stop guessing and start looking at what people actually search for. And it's a story about a mistake we made that cost us three weeks of work.

If you run a business that depends on being found online — which is most businesses in 2026 — this matters. Because the businesses that understand what their customers actually search for are the ones that win the click. The ones that guess are the ones that wonder why their website gets no traffic.

Why We Did It

We were working with a client in the home improvement sector. They'd been doing SEO for two years. They had a blog. They had keywords. They had a "strategy." But their organic traffic was flat. They were getting 2,000 visitors per month and had been stuck there for 18 months.

Their SEO agency had given them a list of 150 keywords. "Target these," they said. The client did. Nothing happened.

We suspected the problem wasn't effort. It was coverage. 150 keywords in a sector with hundreds of thousands of searches per month is like fishing with a single rod in the Atlantic. You might catch something. But you're missing almost everything.

So we decided to map the entire ocean.

How We Mapped 89,000 Search Terms

The Old Way (What Most Agencies Do)

1. Log into Ahrefs or SEMrush
2. Type in a seed keyword like "fireplaces"
3. Export the top 500 related keywords
4. Hand the list to the client
5. Charge £2,000 and call it "keyword research"

This finds the obvious keywords. The ones everyone already knows about. It misses the long-tail, the questions, the local variations, the seasonal trends, and the emerging terms that your competitors haven't found yet.

Our Way (What We Actually Did)

Step 1: We built a data pipeline.

We used a combination of:

• Google Keyword Planner — for search volume data
• Ahrefs — for keyword difficulty and competitor analysis
• Google Search Console — for terms the client was already ranking for
• Custom Python scripts — to aggregate, clean, and categorise the data
• Google Vertex AI — to cluster related terms and identify patterns humans would miss

Step 2: We expanded beyond the obvious.

We didn't just search for "fireplaces." We searched for:

• "fireplaces" + every UK city and county ("fireplaces London", "fireplaces Manchester")
• "fireplaces" + every material ("marble fireplaces", "stone fireplaces", "wood fireplaces")
• "fireplaces" + every style ("modern fireplaces", "traditional fireplaces", "contemporary fireplaces")
• "fireplaces" + every use case ("fireplaces for small rooms", "fireplaces for apartments")
• Question variations ("how much do fireplaces cost", "are fireplaces energy efficient")
• Comparison terms ("fireplaces vs stoves", "gas vs electric fireplaces")
• Brand + generic ("best fireplace brands UK", "cheapest fireplaces")

Step 3: We let AI find the patterns.

This is where it gets interesting. We fed all 89,000 terms into Google Vertex AI and asked it to cluster them by:

• Search intent (informational, navigational, transactional)
• Topic category (product, location, comparison, how-to)
• Seasonality (which terms peak in winter vs. summer)
• Competition level (which terms were underserved)
• Content gap (which terms had weak existing content)

The AI found clusters we never would have spotted manually. For example:

• A cluster of 340 terms related to "fireplace installation regulations" — high intent, low competition, almost no good content
• A cluster of 1,200 terms combining locations with specific fireplace types — "marble fireplaces Surrey" had 90 searches per month with zero dedicated pages
• A seasonal cluster of 2,800 terms that peaked in October–November — perfect for content calendar planning

Step 4: We prioritised by opportunity.

Not all 89,000 terms were worth targeting. We scored each term on:

• Search volume (how many people search it)
• Keyword difficulty (how hard it is to rank)
• Current ranking (if the client was already on page 2, we could get to page 1 faster)
• Content gap (if the existing results were weak, we could win quickly)
• Revenue potential (if someone searching this term was likely to buy)

This gave us a prioritised list of 2,400 terms that represented the highest opportunity.

What We Found (The Interesting Bits)

Finding 1: 94% of Searches Were Long-Tail

Of the 89,000 terms, only 6% had more than 1,000 monthly searches. The other 94% were long-tail — specific phrases with 10–500 searches per month.

Most businesses ignore these because the volume looks small. But collectively, the long-tail terms represented 78% of total search volume. And they had much lower competition. A page targeting "modern marble fireplaces for small living rooms" had almost no competition. A page targeting "fireplaces" had hundreds.

What this means: Don't just chase high-volume keywords. The money is in the long-tail.

Finding 2: Location + Product Combinations Were Massively Underserved

We found 8,400 terms that combined a location with a specific product type. "Gas fireplaces Croydon", "electric fireplaces Bromley", "marble fireplaces Kingston."

The client's existing agency had never created location-specific content. They had one generic "Fireplaces" page. We recommended creating 47 location-specific landing pages. Each one targeted 150–200 terms.

Result: Within 4 months, the client was ranking on page 1 for 89 location-specific terms they'd never targeted before. Organic traffic increased from 2,000 to 6,700 per month.

Finding 3: Question-Based Content Had the Biggest Content Gaps

We identified 12,000 question-based terms ("how to...", "what is...", "can I..."). We analysed the top-ranking content for each. In 67% of cases, the existing content was thin, outdated, or didn't actually answer the question.

This is where AI content production becomes powerful. We built a system that:

1. Identified the top 500 unanswered questions
2. Generated comprehensive answers using AI
3. Had our team edit, fact-check, and add real examples
4. Published 12 question-answering pages per week

Result: The client captured 340 featured snippets within 6 months. Featured snippets sit at position 0 — above the #1 organic result. Click-through rates increased 45%.

Finding 4: Seasonal Terms Were Predictable and Profitable

The AI identified 2,800 terms that showed clear seasonal patterns. Fireplace-related searches peaked in October–November. Outdoor heating searches peaked in March–April. Maintenance searches peaked in September.

Most businesses create content reactively. "It's November, we should write about fireplaces." By then, it's too late. Google takes 2–3 months to rank new content.

We built a content calendar 6 months ahead. Articles were published in August for November peak traffic. In March for June peak traffic.

Result: The client captured peak traffic every season instead of missing it.

What We Got Wrong

The Three-Week Mistake

We were so excited about the 89,000 terms that we tried to create content for all of them at once. We built an automated content pipeline that generated 200 articles per week. We published them as fast as they were written.

Google penalised the site. Not because the content was AI-generated — it was because the content was thin. We were publishing 200 articles but only 30 of them were genuinely useful. The rest were slight variations of the same article with different location names.

What we learned:

• Quality beats quantity. Every. Single. Time.
• AI can identify opportunities, but humans must decide which ones are worth pursuing
• 50 excellent articles beat 500 mediocre ones
• Google rewards depth, not breadth

We deleted 170 of the 200 articles. We kept the 30 that were genuinely useful. We spent three weeks rewriting them with more depth, more examples, and more originality. Traffic recovered within 6 weeks and then exceeded previous levels.

Lesson: AI finds the opportunities. Humans decide which ones matter. Never publish at scale without human review.

What This Means for Your Business

You don't need to map 89,000 terms. Most businesses need 500–2,000 to find their highest-opportunity keywords. But you do need to go deeper than the obvious.

The framework we use now:

1. Map your universe — Use tools + AI to find all relevant terms (aim for 1,000–5,000)
2. Cluster by intent — Group by what the searcher wants (information, product, location)
3. Find the gaps — Identify where existing content is weak
4. Prioritise by opportunity — Score by volume, difficulty, and revenue potential
5. Create depth, not breadth — 50 excellent articles beat 500 thin ones
6. Plan seasonally — Publish 2–3 months before peak demand
7. Measure and adjust — Track rankings, traffic, and conversions monthly

"We mapped 89,000 terms and learned that the real opportunity wasn't in the obvious keywords. It was in the 12,000 questions nobody was answering properly. The businesses that win at SEO in 2026 aren't the ones with the most content. They're the ones with the most helpful content."

Frequently Asked Questions

Jump to: How did you map 89,000 terms? · What did you find? · What went wrong? · Can I do this myself? · How long does it take?

Last updated: 2 July 2026
Prices and tools referenced are current as of this date. AI technology changes quickly, so always check current pricing before committing.