I once exported about 4,300 keywords from a research tool, sorted ascending by difficulty, and started reading. The first 900-odd rows all carried the same score — not similar scores, the identical number, over and over, on phrases with nothing in common except being long. One had a page one made of four Reddit threads and a dead forum. Another had Wikipedia, Investopedia, and three sites with more backlinks than I will ever have. Same score. Same bucket.
That afternoon is when I stopped treating difficulty as a measurement and started treating it as a rough sort order. Finding low competition keywords is a research method you run with your eyes, not a filter you apply in a spreadsheet. This article is that method.
Quick answer: how do you find low competition keywords?
Pull a large set of long, specific phrases from places where people actually ask questions — Google autocomplete, People Also Ask, Reddit, and your own Search Console impressions — then open the search results for each one and read page one manually. A keyword is genuinely low competition when page one is forums, thin aggregators, off-topic pages, or nobody directly answering the question, and when at least one ranking site is no bigger than yours. Validate that real people phrase it that way and that the result page still leaves clicks on the table, then write the format page one is missing.
Why a low difficulty score is not the same as low competition
Every tool computes difficulty differently, but underneath they mostly do one thing: look at the link profiles of the pages currently ranking and turn that into a 0–100 number. Useful signal. Narrow one.
Here is what that number cannot see:
- Whether the ranking pages answer the query. Ten strong domains ranking with pages that miss the intent is an easier SERP than five weak pages that nail it.
- Whether the intent is served at all. The tool sees ten results and computes a score. It does not see that all ten are wrong.
- Freshness debt. A page from 2021 holding position two on a topic that changed last year is a vacancy with a tenant who hasn't moved out yet.
- Whether Google is confused. Forums and Q&A threads on page one mean Google could not find a good article. It is the loudest buy signal in keyword research, and no difficulty score reflects it.
- Who you are. Difficulty is computed for a generic site. You are a specific domain with a specific history in a specific topic.
The bucketing problem, stated honestly
This is the part I wish someone had told me earlier. Difficulty scores in large exports are frequently coarse — bucketed, rounded, or floored. Pull thousands of long-tail terms and you will routinely get hundreds of rows sharing the exact same value, because those terms have thin or missing SERP data behind them and the model falls back to a default rather than admitting uncertainty.
The practical consequence: within a bucket, the score carries almost no information. If 600 keywords all score 9, sorting by that 9 is sorting noise. The real spread between them — some free wins, some unwinnable — lives entirely in the search results page, which you have to go look at.
I still use difficulty. I use it as a first pass to throw away the obviously brutal terms and shrink 4,000 rows to 300. Then the number's job is done.
Proxies that actually predict low competition keywords
These are the signals I check instead. None require a paid tool. All require you to open the SERP.
Phrase length and modifier specificity
Longer phrases are usually easier, but the reason matters more than the rule. A five-word phrase is easier because it encodes a constraint — a role, a use case, a version, a budget, a failure mode. Constraints are exactly what generic pages cannot satisfy without becoming worse for their main query.
The modifiers I look for: for [specific role], without [common requirement], after [event], vs between two narrow options, error strings, version numbers, and "is X worth it for Y". Length alone means nothing — "how to find the best low competition keywords for a blog" is long and still competitive, because it is the same generic intent with padding.
SERP composition — read who is actually there
Open the query in an incognito window and classify all ten results. I score each SERP on five things:
- Forums and UGC on page one. Reddit, Quora, Stack Overflow, niche forums. Two or more is a strong signal — Google is showing threads because it has nothing better.
- Format mismatch. The query asks "how much does X cost" and page one shows product and category pages. Nobody wrote the answer.
- Thin or scraped pages. Aggregators, directory stubs, auto-generated tag pages, 300-word posts that restate the question. They rank because there is no competition, not because they are good.
- Age. If four of the top ten are more than three years old on a topic that moves, freshness alone is a lever.
- At least one site your size. The most decisive check. If every result is a household-name domain, the SERP is stating its entry requirement. If one result is a small independent site, the door is open.
Unserved intent — the highest-value signal
The best low competition keywords are not the ones where weak pages rank. They are the ones where no page exists. You find them by reading the top results and asking a blunt question: after reading these, would the searcher have what they came for?
When the answer is no — the top result answers a broader question and leaves the specific one hanging, the comparison page lists features but never says which to pick, the tutorial stops right before the hard part — you have found something worth writing. That kind of win also holds longer, because you are not out-optimizing a competitor, you are filling a hole.
Can you rank? Competition is relative, not absolute
Difficulty is a property of the keyword. Winnability is a property of the keyword and your site. Two checks settle it.
The weakest-link check. Find the weakest result in the top five — smallest site, thinnest page, oldest content. Ask honestly: can I publish something clearly better than that specific page, on the thing this query rewards? Not better than the top result. Better than the weakest one holding a spot you would be happy with. If you cannot beat the fifth result, the keyword is not low competition for you, whatever the score says.
The topical fit check. Google gives a site latitude where it already has a track record. A keyword three steps outside anything you have published is harder for you than the number implies; one adjacent to a post already collecting impressions is easier. Check Search Console — if your domain already has impressions for related queries, you are not starting cold.
A useful reframe — a keyword at difficulty 30 inside your established topic is often easier for you than difficulty 5 in a topic you have never touched.
A worked shortlist of low competition keywords
Here is the shape of what survives. The patterns are the point; substitute your niche's specifics.
| Candidate keyword pattern | Volume band | Why it is genuinely low competition | Format that wins it |
|---|---|---|---|
[exact error message] [tool name] |
50–200/mo | Page one is a GitHub issue, an unanswered Stack Overflow thread, and two scraper sites. Nobody wrote the fix as a page. | Short diagnostic post: symptom → cause → fix → how to verify it worked |
is [tool] worth it for [narrow use case] |
100–400/mo | Top results answer the general "is it worth it" and ignore the modifier entirely. The intent is unserved. | First-person verdict up top, then the tradeoff table and who should skip it |
[tool A] vs [tool B] for [specific job] |
100–500/mo | Only vendor pages plus one roundup from three years ago. Nobody commits to a recommendation. | Comparison with an explicit decision rule, plus who each option is wrong for |
[niche software] pricing explained |
200–800/mo | Vendor pricing page and directory stubs. Tiers get listed, never translated into real cost. | Worked scenarios with the arithmetic shown at three usage levels |
[process] checklist for [role] |
100–300/mo | Page one is Reddit threads and a conference PDF. No scannable web version exists. | An actual checklist, each item with one line of rationale |
how to [task] without [common prerequisite] |
50–250/mo | Every ranking page assumes the prerequisite. The constraint is the whole query and it is ignored. | Step-by-step for the constrained path, with the limitations stated honestly |
Notice what these have in common. None were chosen because a tool said 4. They were chosen because page one has an obvious defect, and the defect has a name.
The step-by-step process you can run today
Free tools only. Budget two to three hours for your first pass.
Step 1 — Harvest seeds where questions get asked. Google autocomplete (type your topic plus each letter a–z), People Also Ask boxes (expand a few; they branch endlessly), Reddit and niche forum search sorted by recent, YouTube autocomplete, and your site's internal search logs.
Step 2 — Mine your own Search Console. The highest-yield free source, and most people skip it. Filter queries to positions 8–30 with impressions above zero and clicks near zero. Each one is a keyword Google already associates with you, sitting just outside visibility — proven topical fit plus proven demand.
Step 3 — Get rough volume, not exact volume. Keyword Planner gives ranges without an active campaign; that is enough. Trends gives direction, which matters more than the absolute number. A term at 150/mo and climbing beats one at 900/mo and decaying.
Step 4 — Cut hard on obvious difficulty. Use whatever difficulty data you have to delete the top of the range, then stop looking at the number. You are shrinking the list for manual review, nothing more.
Step 5 — Read page one for every survivor. Incognito, and if you can, from the country you target. Score each SERP on the five checks above. Keep only queries that hit at least two, and always require the "at least one site my size" check.
Step 6 — Run the weakest-link test. Write one sentence describing how your page beats the weakest top-five result. If you cannot write the sentence, cut the keyword. This kills about half my list and saves far more time than it costs.
Step 7 — Validate before writing. Details below.
Step 8 — Publish, then watch positions for eight to twelve weeks. Long-tail pages on smaller sites often surface slowly, drift around position 20, then jump. Judging at week two gives the wrong conclusion.
Steps 1 through 5 are where the manual work concentrates, and they are also where this breaks down at scale — reading 300 SERPs by hand is fine once and unbearable monthly. When I need to run this across a whole site and then watch which pages actually move afterward, I use MisarSEO, an SEO research and tracking tool, to keep the candidate list and the position history in one place instead of a spreadsheet I stop updating by week three. The judgment calls in steps 5 and 6 stay manual either way.
Validate before you write
Three checks, ten minutes total, before a keyword earns a draft.
- Does a human phrase it this way? Find the Reddit thread, the forum post, the support ticket. If nobody has ever asked it in those words, it may be a tool artifact — a permutation that exists in a database and not in anyone's head.
- Are there clicks left? Search it. If an AI overview and a featured snippet fully resolve the question in two lines, ranking first may still return very little. Prefer queries whose answers need context, judgment, or steps.
- Does it lead anywhere? A keyword can be easy, real, and commercially pointless. Ask what the reader does next, and whether that path exists on your site.
One last habit: keep a kill list. Write down every keyword you rejected and why. Six months later, when your site has more authority and the SERP has aged, a chunk of that list is newly winnable and the research is already done.
What I stopped doing
- Sorting exports by difficulty and writing the top 20. Inside a bucket, that ordering is noise.
- Treating volume as the tiebreaker. Winnability first, volume second.
- Chasing terms where every result was a major brand, because a tool said 12.
- Judging a new page before eight weeks.
- Writing a longer version of the top result. If the existing page is already right, length is not a strategy.
The method is not complicated: read the search results, be honest about your own site, and let the difficulty score do the one job it is good at — shrinking the list before you start looking properly.
FAQ
What difficulty score counts as a low competition keyword?
There is no universal threshold, and chasing one is the mistake. Under 20 is a reasonable place to begin a manual review, but the number is often bucketed, so hundreds of terms in your export will share a score while having wildly different search results. Use it to shrink the list, then decide from page one.
Are low competition keywords always low volume?
Individually, usually — many of the best sit between 50 and 500 searches a month. The value comes from the aggregate and from intent. Thirty pages at 200/mo each, ranking well and pulling in people with a specific problem, generally outperform one page fighting for a 10,000/mo head term it will never reach.
How long does it take to rank for a low competition keyword?
On a new or small site, expect eight to twelve weeks before the position stabilizes, sometimes longer. Established sites with topical authority in the area often see movement in two to four weeks. The pattern is usually a slow drift followed by a jump, so early positions tell you very little.
Can I find low competition keywords without a paid tool?
Yes, and the free path is stronger than people assume. Search Console for queries you already show up for, Keyword Planner for volume ranges, Trends for direction, autocomplete and People Also Ask for seeds, Reddit for real phrasing, and manual SERP reading for the decision. Paid tools buy speed and scale, not judgment — the judgment is still you, reading page one.
The tradeoff is time. Doing this by hand across fifty candidates costs an afternoon; across five hundred it stops being feasible, and that is the point where paying for automation starts to make sense. Start free, learn what a winnable SERP looks like, and only then decide whether you need the speed.
Top comments (0)