How I use Google Search
Console and Claude to
rank higher every week.
No paid tools. No guesswork. Export your GSC data, feed it to Claude, find the pages with 500 impressions and zero clicks, fix the titles and meta descriptions, request re-indexing, and repeat. This is the exact workflow I ran on forgevertical.com this week — with real data and real results.
Neil Patel built Kissmetrics to millions of visitors and documented the lesson clearly: writing about "marketing funnels" gets you traffic that doesn't convert. Writing about "how to get my first 1000 email subscribers" gets you traffic that buys. The difference is intent. One term is browsed; the other is searched by someone ready to act.
The same principle applies to every article on your site. If you are writing about "AI" you are competing with every AI company, every journalist, and every researcher on the planet. If you are writing about "why my Cape Town business doesn't appear in ChatGPT results" you are competing with almost nobody — and the person searching that exact phrase has a problem they want solved today.
Google Search Console tells you what people are actually searching when they find your site. Claude turns that raw data into a specific action plan. Together they form a weekly loop that compounds over time. Here is the exact process.
Step 1 — Export your GSC data the right way
Analyse this Google Search Console data and give me three lists:
1. High impression / zero click opportunities — queries with more than 5 impressions and 0 clicks, sorted by impressions descending. These are pages Google is showing but nobody is clicking.
2. Position 5–15 with impressions — queries where I am ranking page 1 or early page 2 but not top 3. These are the easiest wins to push higher.
3. My best performing queries with CTR above 5% — so I know what titles and descriptions are actually working and can replicate that pattern.
For each opportunity in list 1, suggest a new title and meta description using long-tail keyword principles — specific, intent-driven, 50-60 characters for title and 150-160 for meta.
What my GSC data actually showed — a real example
// Chapter 02 — Real numbers from forgevertical.comRunning this workflow on forgevertical.com this week produced immediate, actionable findings. Here is what the data showed — real numbers, not hypothetical.
| Page | Impressions | Clicks | Avg Position | Action |
|---|---|---|---|---|
| chris-titus-tech.html | 810 | 0 | 4.73 | Fix title + meta |
| cape-town-tech.html | 183 | 2 | 17.3 | Internal links needed |
| fable5-here-to-stay.html | 172 | 7 | 8.56 | Add Fable 5.1 to title |
| ai-future-of-work.html | 83 | 4 | 12.1 | Longer-tail H2s |
| cape-town-ai-city.html | 20 | 3 | 4.35 | Best CTR (15%) — replicate |
The chris-titus-tech article was the headline finding: 810 impressions, position 4.73, zero clicks. Google is ranking this article on page 1, average position 5, for "chris titus" and related terms — and nobody is clicking. That is a title and meta description problem, not a ranking problem.
The fix was specific. The original title described what the article was about. The new title leads with what people are searching for: "WinUtil by Chris Titus Tech: 30 Million Runs, 48K GitHub Stars". That title contains the tool name, two specific numbers, and signals immediately that this is about his actual utility — which is what people searching "chris titus" actually want.
The Neil Patel long-tail principle in practice
// Chapter 03 — Specific beats broad, every timePatel's Kissmetrics lesson — that "how marketing funnels work" massively outperformed "marketing funnels" — translates directly to every page title you write. The pattern is: broad term → specific question or specific outcome.
| Broad (losing) | Long-tail (winning) | Why |
|---|---|---|
| web design Cape Town | website with llms.txt and AI indexing Cape Town | Specific problem + location = buyer intent |
| GEO agency | how to get my business cited by ChatGPT South Africa | Question format = someone who wants to act |
| password manager | NordPass vs browser passwords — which is safer in 2026 | Comparison = high intent, ready to decide |
| AI job loss | 491 people lost their job to AI today — what to do about it | Specific number + urgency = emotional click trigger |
| Cape Town tech | why Cape Town is Africa's number one tech city in 2026 | Superlative + year + geography = searchable claim |
Step 2 — Fix, request indexing, and set the weekly loop
// Chapter 04 — The feedback loop that compounds
Once Claude gives you the list and the rewritten titles and meta descriptions, the fixes take about 20 minutes per article. Update the <title> tag. Update the <meta name="description">. If you can, update the H1 and the first H2 to match the new long-tail direction. Push to GitHub. Then:
What Claude is actually good at in this workflow
// Chapter 05 — Where AI earns its placeClaude does not replace the judgment — you still decide which articles matter, which keywords are worth targeting, and what your audience actually needs. What Claude does is compress the analysis time from hours to minutes and produce specific, usable output rather than generic advice.
The prompts that produce the best SEO output from Claude are specific about format. Ask for a table. Ask for a character count. Ask for alternatives. Ask it to apply a named principle (Patel's long-tail, SERP intent matching) so it calibrates to a standard rather than making general suggestions. The more specific your prompt, the more specific and useful the output.
Then upload Pages.csv → "Which pages have the worst impression-to-click ratio? Which have the best? What is different about the best-performing titles?"
This workflow is free if you are using Claude.ai. It costs a few rands in API tokens if you are using Claude Code or the API directly. Either way it costs less than any SEO tool subscription — and it uses your actual data rather than estimated search volumes.
The stack behind Forge Vertical
Affiliate links — helps keep this site running at no cost to you.
