Case Study
We compared generic AI keyword ideas vs GSC-based ideas
A SaaS team wanted to know whether AI-generated keyword ideas were enough to guide their content strategy. We compared a generic AI keyword list against ideas generated from their own Google Search Console data inside SEO Autopilot. The difference was not just volume. It was relevance, ranking probability, and business fit.
The question: are generic AI keyword ideas good enough?
The client was a B2B SaaS company with an active blog and access to Google Search Console, but the team was still using generic AI prompts to brainstorm article ideas.
The prompts produced long lists quickly. The problem was that many ideas sounded reasonable but were either too broad, too competitive, too disconnected from the product, or already covered by existing pages.
We ran a simple comparison: 100 generic AI keyword ideas versus 100 ideas generated from the client’s own Google Search Console data through SEO Autopilot.
| Source | How ideas were generated | Main limitation |
|---|---|---|
| Generic AI list | Prompted from product category and target audience | No knowledge of existing impressions, pages, or ranking history |
| GSC-based list | Generated from real Search Console queries, pages, impressions, CTR, and average position | Requires a connected site with enough search data |
Step 1: We generated the generic AI keyword list
First, we asked for article ideas based on the company’s product category, audience, and general positioning. The output looked useful at a glance: clean titles, familiar topics, and a lot of ideas the team recognized from competitor blogs.
But once we scored the ideas, the weakness became clear. Many were category-level topics that large competitors already owned. Others were too far from the product or repeated content the site had already published.
Common problems in the generic AI list
- Ideas sounded plausible but had no evidence of demand for this specific site.
- Several topics were already covered by existing articles.
- Many keywords were dominated by high-authority competitors.
- Some topics attracted informational traffic with little product relevance.
- The list did not distinguish between refreshes, new pages, comparisons, FAQs, or product-led guides.
Step 2: We generated ideas from Google Search Console data
Next, we connected Google Search Console to SEO Autopilot. The tool analyzed real queries the site was already appearing for, along with the pages receiving impressions, clicks, CTR, and average position.
This changed the quality of the ideas. Instead of starting from what a generic model assumed the market might search, the list started from searches Google had already tested against the client’s domain.
SEO Autopilot clustered query patterns, separated brand and non-brand demand, recommended page types, and flagged whether an opportunity should become a new article, a refresh, a FAQ update, or a comparison page.
| GSC signal | How it shaped the idea |
|---|---|
| High impressions, low CTR | Refresh title, metadata, and article angle |
| Average position 8-25 | Prioritize pages already close to ranking |
| Queries landing on the wrong page | Create a dedicated article or comparison page |
| Repeated long-tail query patterns | Group into clusters and build supporting content |
Step 3: We scored both lists with the same criteria
To make the comparison fair, we scored both lists using the same evaluation criteria. We were not judging whether an idea sounded good. We were judging whether the team should actually spend time creating or updating a page for it.
Each idea was reviewed for relevance, ranking probability, search intent, product fit, duplication risk, and whether it belonged in the next 90-day roadmap.
| Evaluation criteria | Generic AI ideas | GSC-based ideas |
|---|---|---|
| Relevant to product positioning | 46 / 100 | 78 / 100 |
| Not already covered by existing pages | 39 / 100 | 71 / 100 |
| Realistic ranking opportunity | 18 / 100 | 63 / 100 |
| Clear recommended page type | 31 / 100 | 84 / 100 |
| Selected for the roadmap | 12 / 100 | 41 / 100 |
What the generic AI list got right
The generic AI list was not useless. It was helpful for broad brainstorming, especially when the team needed to think through common category questions or competitor-style topics.
The problem was that it did not know the site’s actual search footprint. It could not tell which topics already had impressions, which articles were underperforming, which pages were cannibalizing each other, or which ideas were realistic for the site’s current authority.
Useful for ideation
Generic AI produced fast topic directions and helped uncover common themes in the market.
Weak for prioritization
It could not tell which ideas were already validated by the site’s own search data.
What the GSC-based list did better
The GSC-based list was stronger because it started with evidence. If Google was already showing the site for a query, even at a low position, that query became a clue.
SEO Autopilot used those clues to recommend whether to refresh an existing page, create a dedicated article, add FAQs, build a comparison page, or strengthen a cluster.
Why the GSC-based ideas were stronger
- They came from searches the site was already appearing for.
- They revealed gaps between existing pages and user intent.
- They helped prioritize refreshes before new content.
- They reduced duplicate article ideas.
- They produced clearer page-type recommendations.
- They were easier to connect to internal links and existing clusters.
The roadmap we built from the comparison
The final roadmap used both sources, but not equally. Generic AI ideas helped with framing and angle exploration. GSC-based ideas drove the actual priorities.
Out of the final 53 roadmap items, 41 came from GSC-based opportunities and 12 came from generic AI brainstorming that still passed the scoring process.
| Roadmap category | Items selected | Primary source |
|---|---|---|
| Existing article refreshes | 18 | GSC-based |
| New product-led articles | 17 | GSC-based |
| Comparison and alternatives pages | 8 | Mixed |
| FAQ and support-style updates | 6 | GSC-based |
| Exploratory thought-leadership articles | 4 | Generic AI brainstorming |
What made this work
The key lesson was not that generic AI ideas are bad. The lesson was that generic ideas need grounding. Without site-specific data, AI can create plausible content plans that ignore ranking probability, existing page coverage, and actual search behavior.
- Generic AI was useful for brainstorming, but weak for prioritization.
- GSC data revealed opportunities already tested by Google.
- Search Console queries helped separate refreshes from net-new articles.
- Page-type recommendations made the roadmap more actionable.
- The strongest plan combined AI speed with site-specific search evidence.
Generic AI can suggest what a market might care about. GSC data shows what Google is already testing your site for.
Final takeaway
AI can produce keyword ideas quickly, but speed alone does not make a content strategy. The best ideas are not just plausible. They are relevant to the site, realistic to rank for, connected to existing pages, and useful to the business.
SEO Autopilot helped turn Google Search Console data into a better keyword engine. The result was a roadmap built from real search signals instead of generic brainstorming alone.
Build keyword ideas from real search data
Connect Google Search Console, discover proven opportunities, and generate an SEO roadmap based on what your site already appears for.
Generate GSC-based ideas