When you don't have a paid keyword tool, Google Autocomplete looks like a free substitute. Type a starting phrase ("metal roof," "solar shingles"), collect what Google suggests, and repeat with dozens of starting phrases. A phrase that shows up for many different starting phrases seems popular.
We used this method to plan content for a new roofing site. We collected suggestions from many starting phrases and ranked each keyword by how many of them suggested it. We called that its "seed hits."
Then we got access to a keyword tool and checked real monthly search volume. The list was close to upside down.
The data
Same keywords, US search volume from Ubersuggest, September 2026:
| Keyword | Seed hits | Real monthly searches | Difficulty |
|---|---|---|---|
| are solar shingles worth it | 75 (ranked #1) | 70 | 31 |
| will insurance cover a metal roof | 60 (ranked #3) | 40 | 9 |
| solar shingles vs solar panels | 36 | 480 | 33 |
| how long does a metal roof last | not in our top list | 3,600 | 18 |
| metal roof cost | 6 | 12,100 | 28 |
| metal roof vs solar shingles | about 0 | 0 | 4 |
Our #1 keyword gets 70 searches a month. A keyword near the bottom of our list, "metal roof cost," gets 12,100, about 170 times more. "How long does a metal roof last" didn't even make our top list, and it gets 3,600.
If we had built pages in seed-hit order, the first months of content would have gone to the smallest opportunities.
Why it happens
Autocomplete predicts what you're about to type. It isn't a ranking of what people search most.
- Long questions get many suggestions because people start typing them and stop halfway. "Are solar shingles..." can be completed from many angles, so it appears under many starting phrases.
- Short, popular searches get few suggestions because people type them in full. Nobody needs help finishing "metal roof cost." So it rarely appears as a suggestion under other starting phrases, even though thousands of people search it every month.
So seed hits measure how central a phrase is among long-tail questions. That pushes the biggest keywords toward the bottom, often by a factor of 10 or more.
What Autocomplete is good for
We still use it, just not for ranking. It's excellent for:
- Proving a phrase is real. If Google suggests it, real people type it. That's better evidence than a keyword tool's guess for very small terms.
- Collecting FAQ questions. The long questions Autocomplete loves are exactly what belongs in an FAQ section or a "common questions" block on a page.
- Finding the wording people use. "Will insurance cover a metal roof" tells you how homeowners phrase the worry, which helps headlines and copy.
What to do instead for ranking
If you're planning content without a paid tool, or checking a list someone else built:
- Check the short, obvious keywords first. The terms that are missing or low on an Autocomplete-based list ("cost," "how long," "vs") are often the biggest ones. Check those before anything at the top of the list.
- Rank by searches divided by difficulty. A 3,600-search keyword with difficulty 18 is a far better first page than a 70-search keyword with difficulty 31.
- Look at cost per click for value. CPC shows what advertisers will pay for a visitor, which hints at how close the searcher is to buying. In the same research set, one keyword with 40 searches a month had a CPC of $44.98, while a 480-search keyword sat at $8.39. Per visitor, the small one may be worth more. Volume isn't everything.
Check your existing plans
If any of your sites were planned from Autocomplete counts, the same problem is probably sitting in them. We found it on our own sites after this test.
The quick audit takes an hour. Take your top 20 planned topics, add the 10 most obvious short searches in the niche, and check real volume for all 30. If the obvious ones beat your top picks, reorder before you write another page.
