Before you commit months of writing to a niche, you need a defensible reason to believe people actually search for what you plan to cover. Volume tools alone will not give you that reason. This is the cheap, reproducible workflow I use: scrape autocomplete, score the SERP, read the free Keyword Planner range correctly, and then let real Search Console impressions cast the deciding vote.
TL;DR
- A single number from one tool is the weakest signal. Stack qualitative signals (autocomplete depth, “People also ask”, live community threads) against SERP shape (AI Overview present? big publishers locked in?) and only then look at Keyword Planner.
- Free Keyword Planner shows logarithmic ranges (10, 100, 1K–10K), not exact numbers, unless you run a live campaign. Treat the range as a yes/no on demand, not a forecast.
- AI Overviews now sit on the queries that hurt most. As of June 2026, position-one organic CTR drops roughly 58% when an AI Overview is present (Ahrefs), and AI Overview SERPs average an 83% zero-click rate. Score “what is X” questions down accordingly.
- The only ground truth is impressions. Ship two sample articles, request indexing, wait four weeks, and read Search Console. >50 impressions per article = real demand.
Why volume tools mislead indies
Keyword tools round head terms to clean numbers and flatten everything else to zero. That zero is a lie of precision: a query showing “0” in one tool routinely shows 10–100 in another, and a few hundred such long-tail queries collectively out-earn one competitive head term you will never rank for. For a small content site, the signals that actually predict traffic are qualitative — do real humans phrase the question this way, does the SERP serve people like your reader, and can a click still survive the AI Overview. A scorecard makes those judgments explicit instead of leaving them to a number you half-trust.
The five demand signals
Type the head term and check, in order:
| Signal | What “real demand” looks like (June 2026) |
|---|---|
| Google autocomplete | ≥5 related continuations appear; fewer than 2 is a dead topic |
| People also ask | PAA box shows and keeps expanding when you click an entry |
| Live community threads | Reddit / Quora / Stack Exchange answers from the last 12 months |
| Video coverage | YouTube tutorials in the 4–5 figure view range |
| Keyword Planner range | Several related terms land in a non-zero band (10 or 100), not all zero |
Quick verdict: autocomplete + PAA + active community threads on the same topic together mean demand is real. Miss two of those three and the topic is too small or already dead.
Before you start
- Have 30 candidate query phrasings written down. Validating one phrasing in a vacuum is how you misjudge a whole niche.
- Open Google in an incognito window so personalization does not bias autocomplete or the SERP.
- Create a free Google Ads account for Keyword Planner. You do not need to spend or even enter payment details — at account setup, skip the campaign goal and click “Switch to Expert Mode” to reach the full dashboard.
Step by step
1. Scrape Google autocomplete for every candidate
The unofficial suggest endpoint still returns JSON in June 2026. The client=firefox parameter is what gives you the JSON array:
for q in "astro deploy firebase" "firebase hosting cache" "vercel custom domain"; do
echo "=== $q ==="
curl -s "https://suggestqueries.google.com/complete/search?client=firefox&q=$(printf '%s' "$q" | jq -sRr @uri)" \
| jq -r '.[1][]'
sleep 1
doneThe endpoint is undocumented and unsupported, so add the sleep 1 and back off on any HTTP 429/503 — hammering it will get you temporarily blocked. A topic returning fewer than 5 suggestions is suspect; fewer than 2 is dead.
2. Build a SERP signal scorecard
Score 20 candidates into one CSV. These columns capture both demand and winnability:
query,autocomplete_count,paa_present,reddit_recent,youtube_4_digit_views,ai_overview,big_pubs_in_top10,verdict
astro deploy firebase,7,yes,yes,yes,no,2,green
firebase hosting cache,5,yes,yes,no,no,1,green
vercel build failed,9,yes,yes,yes,yes,4,yellow
nextjs middleware example,4,no,no,no,no,8,redGreen = pursue, yellow = pursue carefully, red = drop.
3. Read each SERP’s shape
Search each query in incognito and note the layout. What each element tells you:
PAA box -> demand signal
AI Overview -> click-through risk (see step 4)
Reddit in top 10 -> indie-friendly intent, you can compete
YouTube in top 5 -> consider repurposing as video too
Wikipedia in top 3 -> likely informational only, harder to monetize4. Down-score queries with an AI Overview
This is the signal that changed most. When an AI Overview renders, position-one organic CTR falls about 58% versus the same query without one (Ahrefs, updated late 2025), and AI Overview SERPs average an 83% zero-click rate as of June 2026 versus roughly 60% on plain SERPs. The split that matters: AI Overviews gut broad “what is X” / “how does Y work” traffic, but specific, action-shaped questions (“fix [error string] in [tool] 2026”, “[framework] deploy to [host] step by step”) still earn the click because the searcher needs to do something, not just read a definition. Subtract a point from any candidate where a generated answer sits above the organic results.
5. Check community freshness manually
Spend 30 minutes confirming people still talk about these terms. Use a site: filter so you only see one source at a time:
https://www.google.com/search?q=site%3Areddit.com+%22astro+deploy+firebase%22No Reddit/Quora discussion in the last 12 months usually means the topic is too niche to scale into hundreds of articles.
6. Read the free Keyword Planner range correctly
Sign in to Google Ads → Tools → Keyword Planner → “Get search volume and forecasts” and paste your 20 candidates. Without active ad spend, the “Avg. monthly searches” column shows logarithmic ranges (10, 100, 1K–10K, 10K–100K) — Google reserves exact numbers for accounts running live campaigns (a minimal ~$5–10/day campaign unlocks them). Do not chase the exact figure at validation stage. Read it as a binary: a non-zero band confirms the query exists; the Competition and Top of page bid columns tell you how monetizable and contested it is. Both matter more here than the volume estimate.
7. Score and tag each query
Aim for at least 15 candidates in the green bucket before committing. The rule:
score = (autocomplete >= 5 ? 1 : 0)
+ (paa_present ? 1 : 0)
+ (reddit_recent ? 1 : 0)
+ (youtube_views ? 1 : 0)
+ (ai_overview ? -1 : 0)
+ (big_pubs_top10 < 4 ? 1 : 0)
# score >= 3 -> green
# score 1-2 -> yellow
# score <= 0 -> red8. Ship two samples and let impressions decide
Publish two articles on your strongest green queries. Submit each in Search Console (URL Inspection → Request Indexing) and wait four weeks. Real impressions are the only ground truth that overrules every tool. Pull them through the Search Console API (v1, still on the webmasters/v3 path as of June 2026):
# Search Console API quick check (after OAuth).
# Defaults to 25,000 rows/request; cap is 50,000/day per site per search type.
curl -X POST "https://www.googleapis.com/webmasters/v3/sites/$SITE_URL/searchAnalytics/query" \
-H "Authorization: Bearer $TOKEN" \
-H "Content-Type: application/json" \
--data '{
"startDate": "2026-04-22",
"endDate": "2026-05-22",
"dimensions": ["query"],
"rowLimit": 100
}'If your two samples clear 50 impressions in four weeks, the topic has real demand — go build the niche.
Validation checklist
- 30-query candidate list captured and scored in one CSV.
- Autocomplete scraped for every candidate, not just the head term.
- Reddit / Quora freshness checked manually within the last 12 months.
- Keyword Planner range and Competition column logged per query.
- AI Overview presence recorded and scored down.
- Two sample articles published and submitted for indexing.
After four weeks, you are looking for: non-zero impressions on both samples; average position under 50 on at least one target query (room to climb); and a few queries Search Console surfaced that you never anticipated — that last one is proof the topic has real long-tail width.
Common pitfalls
- Trusting one tool. Ahrefs, Semrush, and Keyword Planner disagree, sometimes by an order of magnitude. Use the range as a tiebreaker, not a verdict.
- Confusing viral with searched. A topic trending on X or in a Reddit spike is not the same as sustained search demand.
- Reading silence as zero. Empty autocomplete sometimes just means Google suppressed suggestions for that string, not that nobody searches it.
- Ignoring AI Overview presence. If the generated answer fully resolves the question, your click-through will be a fraction of what the volume implies.
- Skipping qualitative signals the moment a tool hands you a number.
- Validating in a logged-in Chrome session. Personalization warps both autocomplete and the SERP; always use incognito.
FAQ
- How many monthly searches is “enough”? For an indie content site, 30–100 monthly searches per article is plenty when you have hundreds of articles. Coverage breadth beats per-article volume.
- Should I pay for Ahrefs or Semrush to validate? No. Free Keyword Planner plus autocomplete is enough at this stage. Pay for a research tool only once you have traffic and need to scale keyword discovery.
- Why does Keyword Planner show “10K–100K” instead of an exact number? Google only shows exact average monthly searches to accounts with active ad spend; everyone else sees logarithmic ranges. A small live campaign (~$5–10/day) unlocks exact figures, but you rarely need them to validate.
- Does an AI Overview kill SEO for my topic? It hits top-of-funnel “what is X” queries hardest — roughly a 58% position-one CTR drop and an 83% zero-click rate on those SERPs as of June 2026. Specific, action-oriented questions still earn clicks because the searcher has to act, not just read.
- How do I judge competition from the SERP? Search 10 target queries and count how many top-10 results are big publishers, Reddit, YouTube, or an AI Overview. If big publishers plus video dominate, you will not win that query.
- What if every candidate is yellow, not green? Pick the five yellows with the strongest Reddit signal and the weakest big-publisher coverage, ship samples, and let impressions decide. Yellow with proven impressions is a fine place to build.
Related
- How to pick a niche that has search demand
- Planning a long-tail keyword site from day one
- Submit a website to Google
- Content volume vs quality
- URL inspection tool: how to use
Tags: #Indie dev #Website planning #SEO #Long tail #Search Console