What a gap analysis answers
A gap analysis compares two sets of queries: those the competitors rank for, and those your site ranks for. Each query in the combined list falls into one of three classes, and the classes drive what happens next.
| Class | Meaning | Usual action |
|---|---|---|
| Shared | Your site and competitors all rank on the first results page | Monitor; improve only if the business value is high |
| Weak | Competitors rank on the first page; your site ranks, but further down | Improve the page that owns the query |
| Missing | Competitors rank; your site does not rank at all | Consider a new page, or extend an existing one |
What it does not answer matters as much. It does not say whether a missing query fits your business, whether its intent is one your site should serve, or how much traffic a new page would earn. Those come from the later steps, and the last one never gets an exact answer. It also says nothing about queries nobody ranks for yet, which keyword research from customers' own words still has to find.
The method sits between the wider competitor analysis guide, which covers competitors across the whole marketing mix, and the keyword mapping tutorial, which turns clusters into owner URLs. Its output is rows for the map.
Before you start
- The clustering method. The keyword clustering guide is the prerequisite: the gap list is clustered the same way as the rest of your research.
- The keyword map. Your existing map, so new rows can be checked against existing ones. The keyword map template sets out the columns if you do not have one.
- Search Console access for the site, with enough history to cover seasonality.
- A third-party keyword tool that can export the queries a domain ranks for, with estimated positions, ranking URLs and estimated Search VolumeThe estimated number of times a query is searched in a period, usually per month.. Any tool will do; what matters is using the same tool and the same market settings for every site, and treating its figures as estimates.
- A list of your products, services and topics, so fit can be judged quickly.
- A spreadsheet or database that can join lists of tens of thousands of rows.
The steps
Choose search competitors, not business rivals
List your most important category, product and problem queries, and look at the results page (SERPThe page of results a search engine shows.) for each in the market you serve. Note every domain that appears repeatedly. Some will be the rivals you expect; others will be publishers, marketplaces, review sites and specialists that never appear in a sales conversation but take the clicks you want. The competitor analysis guide calls these intermediaries. Pick the three to five domains that recur most across your priority queries, and include at least one that ranks with the kind of page you could realistically publish. If a domain covers far more than your subject, such as a general marketplace, restrict its export to the relevant section.
Export their ranking queries
In the third-party tool, export each competitor's ranking queries for the same country, language and device, with the query, ranking URL, position and estimated volume. Limit the export to positions on the first two results pages; queries where a competitor sits further down are rarely evidence of anything. Record the date of the export, because rankings move and the tool's snapshot ages quickly. Export your own domain from the same tool with the same settings, so the comparison in step five compares like with like.
Export your own queries
From the Search Console performance report, export queries with clicks, impressions and average position for the same country, over a period long enough to cover seasonal patterns, and the same data by page. The help documentation for the report explains what each measure counts and that some queries are not shown, for privacy reasons, so the export is not every search your site appeared for. Search Console is your source of truth for your own site; the tool's export of your domain is the like-for-like comparison with competitors. Where the two disagree, trust Search Console and note the difference.
Normalise and deduplicate
Combine all exports into one table with a row per query and a column per site. Lower-case and trim every query, remove obvious duplicates, and merge variants that differ only in plurals, word order or accents. Separate out competitors' brand queries into their own list: they are useless as gaps, but the ones combined with "vs" or "alternatives" may be worth a comparison page later. Remove queries outside your market, job and login queries, and anything clearly navigational to a competitor's own product. What remains is the comparable set.
Classify each query as shared, missing or weak
Apply the three classes from the table above with thresholds you set and write down. A workable default: shared if your site and at least one competitor are both on the first page; weak if a competitor is on the first page and you rank lower; missing if at least one competitor ranks on the first two pages and your site has no ranking and no meaningful impressions in Search Console. Count how many competitors rank for each missing query: a query where several competitors rank is stronger evidence than one where a single site does.
Cluster the missing and weak queries
Group the missing and weak queries into clusters using the same method as the rest of your research, by results-page overlap and by meaning, as the clustering guide describes. Most of the list will be long-tail queriesA specific, usually longer query searched rarely on its own but often as a group. that make sense only as members of a cluster. Note each cluster's head term, its member queries, the competitor URLs that rank for it and the type of page they are: guide, category, tool, comparison, forum thread. The cluster, not the query, is the unit for every decision from here on.
Check intent and fit
For each cluster, label the Search IntentThe goal behind a search query: to learn, to find a site, to compare or to act. with the intent classification guide and look at the live results page. Then ask three questions. Does the business offer, or have real expertise in, what the searcher wants? Can the site produce the kind of page that ranks, given the format the results page rewards? Would a visitor from this cluster be someone the business wants? Drop clusters that fail any of the three, and record why, so the same idea is not reconsidered every quarter. Missing clusters that an existing page could cover go to that page's row rather than to a new page.
Size and prioritise
Sum the estimated volume of each surviving cluster to get its relative size, and treat it only as that: this cluster is larger than that one. The SEO traffic estimator turns a volume and a position into an estimate under its own stated assumptions, which is useful for comparing scenarios between clusters and misleading as a forecast. Score each cluster on business value, relative size, the strength of the pages that rank and the effort of the page needed, and order the list. Write the result as candidates with reasons, never as traffic the site will gain.
Add the results to the keyword map
Add each prioritised cluster to the keyword map as a row: cluster, head term, intent, format, relative size, the owner URL (an existing page to improve, or a proposed new one), priority, and "gap analysis" with the date as the source. Before adding a new owner URL, check that no existing row already owns a closely related cluster, because a new page for a query an old page half-covers is how Keyword CannibalisationTwo or more pages on one site competing for the same query and intent, splitting signals. starts. The keyword mapping tutorial covers the decisions for each row; repeat the gap analysis at the quarterly research review with fresh exports.
A gap table (illustrative)
A pet-supplies store compares itself with three category competitors
Nettlecove Pet Supplies, an invented online pet-supplies store, chooses its search competitors from the results pages for 40 priority category queries: a large pet retailer, a pet-care publisher and a specialist aquarium shop appear most often. After normalising, the combined export holds 18,600 queries; 2,100 are shared, 3,400 weak and 6,900 missing, and the rest are competitors' brand queries or out of scope. Clustering the weak and missing queries produces 410 clusters. The intent and fit check drops 260 of them, including veterinary-treatment clusters the store cannot answer with authority and live-animal clusters it does not sell. A section of what remains:
| Cluster (head term) | Class | Competitors ranking | Est. monthly volume (tool) | Ranking page type | Decision |
|---|---|---|---|---|---|
| how to set up a tropical fish tank | Missing | 3 | 5,400 | Guide | New care guide, high priority |
| best food for senior dogs | Weak (position 14) | 2 | 3,600 | Buying guide | Improve existing buying guide |
| rabbit hutch size guide | Missing | 2 | 1,900 | Guide | New care guide |
| cockapoo grooming tools | Missing | 2 | 1,300 | Category and guide | Breed-specific landing page |
| cat water fountain | Shared | 3 | 8,100 | Category | Monitor |
| guinea pig bedding | Weak (position 11) | 3 | 2,400 | Category | Improve category copy and filters |
| dog cooling mat | Missing | 1 | 2,900 | Category | Category exists but is not indexed; fix |
The 150 surviving clusters fall into two patterns the store had not planned for: care guides for setting up and keeping each kind of pet, and breed-specific pages for the dog breeds most searched with product terms. They go into the keyword map as 38 new rows and 52 improvement rows on existing pages, with the remaining clusters attached as secondary queries to those rows.
Reading third-party data with care
Every figure from a third-party keyword tool is a model, not a measurement. The keyword research hub's table of sources says so directly: volumes are estimates, often averaged over a year and grouped with variants, and difficulty scores are each tool's own invention. For a gap analysis, four cautions follow.
- Positions are snapshots. The tool checks rankings on its own schedule, from its own locations, for its own query list. A competitor's position may have changed since, and personalised and local results differ from the tool's view.
- Coverage is incomplete. Tools track a sample of queries. The long tail, where many real gaps sit, is the part least covered, so the absence of a query from an export does not mean nobody searches for it.
- Volumes are relative at best. Use them to compare clusters with each other, never to predict clicks. Real click-through depends on position, the results page's features and the brand, none of which a volume figure knows.
- Tools disagree. Two tools can give very different volumes and positions for the same query. Use one tool consistently within an analysis, and do not mix exports from different tools in one comparison.
Search Console has limits too: it reports only queries where your site appeared, omits some for privacy, and gives an average position across many searches. It is still the only first-hand data in the analysis, and after new pages are published, it, not the tool, is how their performance should be judged.
Common mistakes
- Choosing business rivals instead of search competitors. The sites taking the clicks are often publishers and marketplaces.
- Treating estimated volumes as forecasts. They rank clusters against each other; they do not predict visits.
- Working query by query. Without clustering, the list produces dozens of pages for one need.
- Skipping the fit check. A competitor's rankings for things you do not sell are not your gaps.
- Mixing tools or settings. Different countries, devices or tools in one comparison make every class unreliable.
- Creating new pages for queries an existing page half-covers. Improve the existing page instead, or the two will compete.
- Running it once. Rankings, competitors and results pages change; repeat at each quarterly research review.
