Google Ads Keyword Planner in 2026: Building a Keyword Set You Can Trust

Google Ads Keyword Planner is the most used and most misread tool in the account. It reports ranges instead of numbers, collapses different phrasings into one row, and forecasts clicks from market averages that have nothing to do with your landing page. Then the media plan built “from the Planner” misses reality by two or three times, and the algorithm gets blamed.

This is a working breakdown: what the numbers actually mean, which settings move them the most, how to turn an export into an account structure, and how to calculate — before launch — whether a keyword cluster can ever produce enough data to optimise on. No “pick relevant keywords” advice, just the sequence and the places where money usually leaks.

What Google Ads Keyword Planner really shows

Accept one thing first: the Planner is an order-of-magnitude estimator, not a source of exact data. It was built to help you start, not to replace what your own account will tell you later.

Ranges instead of numbers

“Avg. monthly searches” appears as coarse buckets — 10–100, 100–1K, 1K–10K — for accounts without meaningful ongoing spend. Accounts that spend consistently see tighter values. That is granularity policy, not a glitch.

Practical rule: while you see buckets, never plan on the top of the range and never on the bottom. Use a geometric midpoint and carry the spread through the plan. For a 1K–10K bucket, roughly 3K searches a month is a sane working figure — not 10K.

Close variants get merged

By default the Planner groups close variants: singular and plural, misspellings, word order, sometimes near-synonyms. A single row like “running shoes” may bundle a dozen actual phrasings. That inflates the apparent volume of one exact phrase, and the traffic then “disappears” when you split everything into exact match.

Turn close-variant aggregation off when you need per-phrase accuracy, and never design your account as though each exported row were an independent stream of demand.

The forecast is not a promise

The Forecast tab models clicks and cost from the bid you enter and averaged auction data for the geography you selected. It knows nothing about your Quality Score, ad relevance, landing page or account history. In practice the outcome lands on both sides of the estimate: strong ads and a fast, relevant landing page buy cheaper clicks; weak ones pay a premium.

The right question to ask the Planner is “is there a market here, and how big is it?” — not “how much will I make?”

The settings that change every number on the screen

Lock down the conditions before you look at a single figure. The same phrase can differ several times over across setups.

Setting	Effect	How to set it
Location	Changes volume and competition dramatically	Exactly the regions you will serve in — never “all countries”
Language	Filters the data source, not your future targeting	The language your audience searches in, not your interface language
Search networks	“Google and search partners” inflates volume	Google only for the first pass
Date range	Seasonal niches swing several times over	12 months for the baseline plus the monthly chart for seasonality
Brand filter	Competitor brands bloat “your” demand	Split brand terms out immediately unless you are planning brand defence

Search partners deserve a separate decision, because the setting changes both your estimate and your later reporting. How that network behaves and when it is worth keeping on is covered in the piece on where Google Search Partners actually place your ads.

Five seed sources, not one

Output quality is capped by seed quality. Feed the Planner five obvious phrases and it returns variations of those five phrases. Build the seed from independent sources:

1. Your site and competitor sites. The Planner can generate ideas from a URL. Run your landing page plus three or four competitors and you get several different vocabularies for the same market.
2. The search terms report from live campaigns. The only source of real rather than modelled phrasing. Even a small account produces hundreds of live queries a month.
3. On-site search and support tickets. People type there the same way they type into Google, minus any advertising filter.
4. Product language versus problem language. You sell “sales CRM”; the buyer searches “stop losing website leads”. The second vocabulary is usually wider and cheaper.
5. Adjacent and substitute solutions. People looking for an alternative to a competitor, or for a manual workaround, are demand that pure head-term buyers never reach.

From a keyword list to an account structure

An export is raw material, not a keyword set. The job is turning it into a structure the bidding algorithm can actually learn from:

1. Cut by intent. Research queries (“what is”, “DIY”, “free”, “salary”, “jobs”) drain budget. Many of them are negative keywords before you ever launch.
2. Split by stage. Transactional (“buy”, “price”, “near me”, “quote”), comparative (“vs”, “alternative”, “best”), problem-aware (“how to fix”). Different stages need different ads, landing pages and acceptable CPA.
3. Cluster by meaning, not by string. A cluster is a set of phrases that one landing page and one ad answer honestly. Shared word stems are not the criterion.
4. Check cluster volume. A cluster with 30 monthly searches does not deserve its own ad group — it will never accumulate data and will slow learning down.
5. Assign match types. Phrase and broad under Smart Bidding, exact where you need hard control. The mechanics and the traps are in the breakdown of keyword match types in 2026.

One 2026-specific principle: stop shattering the account into dozens of single-keyword ad groups. Smart Bidding learns from conversions, not from keywords, and fragmentation leaves every group with too little signal to learn from. The connection to learning behaviour is covered in the article on the Smart Bidding learning period and the edits that reset it.

Minimum viable volume: will this cluster ever learn?

The real question for any cluster is whether it can generate enough conversions. That is back-of-envelope maths you do before launch, not after.

Take a cluster with 3,000 monthly searches:

• impression share you realistically capture early on — 30–50% (you are not alone in the auction);
• search CTR in an average commercial niche — a 3–7% ballpark, heavily dependent on position and assets;
• landing page conversion rate — a 1–4% ballpark for cold traffic.

Searches are the ceiling on impressions. At 40% impression share that is roughly 1,200 impressions, at 5% CTR about 60 clicks, at 2% conversion rate one or two conversions a month. That is not a workable cluster for automated bidding: an algorithm wants something in the region of 30 conversions per campaign per month to optimise reliably.

Reverse the maths: 30 conversions at a 2% conversion rate and 5% CTR needs roughly 1,500 clicks and 30,000 impressions — a cluster of about 60,000–75,000 monthly searches at 40–50% impression share. That is why narrow niches live on manual or semi-manual bidding, not because “automation doesn’t work”.

Cluster search volume/mo	Ballpark conversions/mo	What to do with it
under 5K	1–3	Merge with neighbouring clusters; no dedicated campaign
5K–20K	3–12	One ad group inside a shared campaign, manual oversight
20K–60K	12–30	Its own campaign; tCPA once data accumulates
60K and above	30+	Full campaign on automated bidding, segmentation viable

Treat those figures as reference points rather than benchmarks — CTR and conversion rates vary wildly by vertical. The value of the exercise is spotting, in advance, that a cluster will never accumulate signal.

Reading the Forecast tab properly

The forecast is useful for the shape of the curve, not the absolute numbers. As you move the bid, you watch clicks rise and efficiency fall — that is the elasticity of the auction. Three practical uses:

• Find the entry threshold. If forecast clicks collapse to near zero below a certain bid, that auction has a minimum price of participation. Entering below it is theatre.
• Estimate the ceiling. If doubling the bid adds 10% more clicks, demand is capped, and growth has to come from new clusters or new channels rather than from bidding.
• Test it against unit economics. Drop the forecast CPC into your model. If the maths fails at market CPC even with optimistic conversion rates, that is a reason to change the offer or the landing page, not a reason to “test and see”.

Once you are live, two honest in-account tools replace the forecast: bid, budget and target simulators built on your own data and Performance Planner for budget planning. Both model your account history rather than market averages.

Processing the export before you touch the account

A Planner export is 500–5,000 rows, of which maybe a fifth ever reaches the account. A processing order that saves hours:

1. Tag brands. A dedicated column: own brand, competitor brand, non-brand. Three different economics, three different campaigns. Blending them produces a pleasant average CPA that means nothing.
2. Tag intent. Transactional, comparative, informational, navigational. Formula-based tagging on markers (“buy”, “price”, “quote”, “reviews”, “how to”, “what is”, “free”, “DIY”) gets you 80% of the way.
3. Strip obviously foreign demand. Jobs, courses about your topic, second-hand, repair, wholesale phrasing if you sell retail.
4. Pivot volume by cluster. Cluster totals, not individual rows, are what feed the volume calculation.
5. Keep what you deleted. Everything removed in steps 3 and 4 is your starting negative keyword list, not trash.

That spreadsheet later becomes the document that explains where the media plan came from. When someone asks why you are not showing for a particular query, the answer is a row, not a memory.

Keywords in the age of broad match and AI formats

In 2026 keywords increasingly act as an intent signal for the algorithm rather than a hard serving filter. Broad match under Smart Bidding leans on user context, account history and conversion data as much as on the phrase itself. Three consequences for planning:

• Seed width beats tail length. Manually harvesting thousands of long-tail variants used to be the job. The system now finds them — but only if the intent signal and the conversion data you feed it are clean.
• Negatives matter more than keywords. The broader the match, the more the constraints do the steering. A starting negative list is a reach-management tool, not housekeeping.
• Signal quality caps traffic quality. If the algorithm learns on junk conversions, no keyword set will rescue it. Splitting goals correctly is covered in primary vs secondary conversions and lead quality.

Eight mistakes that waste the whole exercise

1. Reading volume without geography. Worldwide volume is impressive and irrelevant to a single-city campaign.
2. Planning on the top of the bucket. Systematically inflates plans by two to three times.
3. Ignoring seasonality. A 12-month average in a seasonal market is a number that occurs in no actual month. The monthly chart answers it in a minute; budgeting around it is covered in seasonality and budget planning.
4. Importing everything the tool suggested. The Planner produces ideas, not an approved list. Half the rows are competitor brands and research intent.
5. Confusing competition with difficulty. The competition column counts how many advertisers bid on a phrase, not how hard it will be for you to win.
6. Skipping negatives at the planning stage. The junk intersections are visible in the export; they are cheaper to cut before the first impression.
7. Treating the Planner as a substitute for the search terms report. Real phrasing only arrives with real traffic.
8. Refreshing keywords once a year. Demand shifts faster: new phrasing, new competitor products, changing audience language.

After launch: where the truth moves

The Planner covers roughly the first two weeks of a campaign’s life. After that your own traffic is the source of truth:

• pull the search terms report weekly, promote working phrasing into the keyword set and push junk into negatives — the working routine is in negative keywords and search terms analysis;
• check which clusters convert and which only collect impressions, and move budget accordingly;
• return to the Planner quarterly with a seed drawn from live queries rather than from memory;
• watch impression share separately: capturing 20% on a profitable cluster means growth is available with no new keywords at all.

If ads collect few impressions despite live demand, the cause is usually not the keyword set — the diagnosis order is in why your ads are not showing and how to diagnose it. And if conversion numbers look wrong, start with diagnosing conversion tracking, or you will be optimising a keyword set against bad data.

Pre-launch checklist

• Location, language, network and date range are fixed and match the campaign plan.
• The seed comes from at least three independent sources.
• Competitor brands are separated and deliberately included or excluded.
• Every cluster has a landing page that honestly answers it.
• Expected monthly conversions are calculated per cluster.
• A starting negative list exists before launch.
• Forecast CPC has been dropped into unit economics and the maths holds.
• A date is set for the first review against actual search terms.

For teams launching regularly this becomes a template. If you would rather start from a ready-made system for keyword work and account structure, our Google Ads agency service and agency account setup follow the same logic, with worked examples and export templates.

FAQ

Why does Keyword Planner show ranges instead of exact numbers?

Data granularity depends on account spend. Accounts without meaningful ongoing spend get bucketed ranges; consistent spenders see tighter values.

Can I use the Planner without running campaigns?

Yes, any account has access, but the volumes will be bucketed and the forecasts averaged. That is enough to size a market, not enough to build a media plan.

How close is the click forecast to reality?

Fine as an order of magnitude, unreliable as a promise. A 1.5–2x gap in either direction is normal, because the forecast cannot see your ad quality or landing page.

What if the niche has almost no volume?

Do not try to manufacture volume with broad match. Realistic options: expand into adjacent and problem-aware demand, add demand-generation channels such as Demand Gen and YouTube, and work your existing customer base. A micro-niche on Search has a ceiling, and you want to know it in advance.

Should I disable close variants in the export?

Leave them on when estimating total demand — it is a more honest picture of the market. Turn them off when designing exact match, or you will overestimate individual phrases.

Is “high” competition bad?

It only means many advertisers have bid on that phrase. High competition usually correlates with high commercial value. The decision comes from unit economics, not from a colour-coded indicator.

How do I handle seasonality at the planning stage?

Read the monthly chart for your key clusters over 12–24 months and budget to the seasonal profile rather than the average. Remember that peaks raise CPC as well as volume.

Is it worth generating keywords from competitor URLs?

Yes — it is the fastest way to borrow a market’s vocabulary. Verify the output, though: the tool pulls irrelevant site sections too, especially from multi-product companies.

How many keywords belong in one ad group?

As many as one ad and one landing page can answer honestly. In practice that is usually 5–20 phrases of the same intent, not one keyword per group and not two hundred mixed ones.

How often should the keyword set be rebuilt?

Full review quarterly, incremental additions from the search terms report weekly. Fast-moving markets need a shorter cycle.

The Planner shows volume but my campaign gets no impressions. Why?

Search volume is demand, not a guarantee of entering the auction. Typical causes: bid too low, budget capped, geography too narrow, ad status issues, poor landing page experience. Diagnose from budget and statuses, not from keywords.