PPC Keyword Research in 2026: A Practitioner's Guide

If you learned PPC keyword research before 2022, half of what you learned now does the opposite of what you want.

The shape of the work changed harder between 2020 and 2026 than it did in the entire decade before. Long exact-match keyword lists used to be the deliverable. Today the deliverable is a small set of themed clusters, a watertight list of negatives, and a smart-bidding strategy that lets the algorithm find queries you'd never have written down.

This is how we actually do PPC keyword research for a Google Ads account today. Not the 2018 playbook with a 2025 paint job.

What's changed since the last time you read a guide on this

Three shifts in the last few years matter before any workflow makes sense.

Broad match isn't a dirty word anymore. Google has loosened the match-type definitions, and broad match paired with Smart Bidding now outperforms tight phrase or exact in most accounts that have enough conversion data to feed the algorithm. Industry research suggests broad-with-smart-bidding can drive meaningfully higher conversion rates and fewer irrelevant matches than manual broad alone. The 2020 advice to "avoid broad match at all costs" is now actively bad in many accounts.

Single-keyword ad groups are dead. The SKAG pattern (one keyword per ad group across exact, phrase, and broad) lost its point when Google expanded close-variant matching across all match types in 2018 and 2019. Themed ad groups, where 3 to 20 related keywords share a single intent, now beat SKAGs because they pool data the algorithm needs to learn faster.

A SKAG is a single-keyword ad group (the old structure). A STAG is a single-theme ad group (the modern one). We'll come back to the structural fix in a section below.

Performance Max ate roughly two thirds of Google ad clicks. Per Google's February 2026 Ads update, Performance Max now drives the majority of clicks on the network. Keyword research now feeds two different beasts: Search campaigns where you still pick keywords, and Performance Max where you signal intent through audience signals, asset themes, and negatives.

The workflow that follows still works for both. You just translate the output differently for each campaign type.

Start with the landing page, not the keyword tool

The single biggest improvement we make on inherited accounts is forcing keyword research to start at the landing page.

Open the page the campaign will point at. Read it like a buyer. List every product term, feature, problem, synonym, and buyer phrase the page actually answers. Most pages give you 10 to 20 seed terms before you've opened a single tool.

This step looks obvious. Almost nobody does it.

If your landing page doesn't deliver on a query, you can't bid on it. That's the cheapest negative-keyword work you'll ever do, because it happens before you spend a penny.

Sort your seeds into rough buckets as you go:

• Product or service terms.What you sell, in the words the page uses.
• Feature or attribute terms."Stainless steel", "next-day delivery", "MRI-safe".
• Problem terms.The pain or job the page solves, in the buyer's vocabulary.
• Synonyms and regional variants."Trainers" vs "sneakers", "lift" vs "elevator".
• Brand and competitor terms.Keep these separate. They go in their own campaign with their own bid strategy.

When the brand and competitor terms live in the same campaign as the generic ones, your bid strategy can't optimise for them properly. Different intent, different conversion rate, different acceptable CPC. Different campaign.

Pull commercial-intent terms with Keyword Planner (and look beyond it)

Google's Keyword Planner is the floor of PPC keyword research, not the ceiling. It's the only tool with Google's own auction data, so the volume and CPC signals are directional in a way no third-party tool can match.

The catch is that Planner shows you volume buckets, not exact figures. Use it for ranking ideas relative to each other, not for absolute search-volume claims. A keyword in the 1K to 10K bucket isn't necessarily ten times bigger than one in the 100 to 1K bucket. It might be twice as big. You won't know.

Where to look beyond Planner:

• The search terms report on existing campaigns.This is the highest-quality keyword source you'll ever have access to, because it shows you queries that actually triggered your ads, not estimates of what people might search. If the account is already live, start here, not in Planner.
• Ahrefs or Semrush keyword tools.Paid, but the traffic potential, parent topic, and SERP-feature signals are useful when you're sizing up a new vertical.
• Reddit and niche forums.Free. Where you find the vernacular the marketing department doesn't use.

Before you analyse intent on any keyword, run a financial-fit screen. If your target CPA is £30 and the landing page converts at 5%, your maximum CPC is £1.50. Anything above that is structurally unprofitable. Cull it from the list before you do anything else with it.

This single screen kills more bad bids than every other technique combined.

Sort by intent, not by volume

The old mental model was brand, generic, related, competitor. That bucket structure still works. The 2026 overlay is commercial-intent strength, ranked from highest to lowest:

1. Transactional."Buy X", "X near me", "X price", "book a demo". Highest priority. Smallest list. Highest CPCs.
2. Commercial investigation."Best X", "X vs Y", "X reviews", "alternatives to Y". Strong intent, bigger list. Often where the spend concentrates.
3. Informational with buyer signals."How to fix X", "X for [specific use case]". Lower intent but defensible if the landing page genuinely teaches and captures.
4. Pure informational."What is PPC". Usually skip for paid search. SEO can have this one.

Build themed ad groups (STAGs), not single-keyword groups (SKAGs)

A themed ad group contains 3 to 20 related keywords that share a single intent. The ads in that group can speak to that one intent specifically. The algorithm sees consolidated data and learns faster.

A single-keyword ad group contains one keyword across exact, phrase, and broad. The intent is the same across the three. The data is fragmented across three line items the algorithm has to learn separately.

Here's what good STAG structure looks like for a B2B SaaS trial-signup campaign:

Campaign: B2B SaaS, Trial Signups
├── Ad group: Demo intent
│     "book a demo", "request demo", "saas demo", "schedule a demo"
├── Ad group: Comparison
│     "alternatives to [competitor]", "[competitor] vs [us]"
└── Ad group: Problem
     "fix [problem]", "[problem] software", "[problem] tool"

The rule of thumb: if you can't write one ad headline that's relevant to every keyword in the group, the group is too broad. Split it.

The other rule of thumb: if you're tempted to put two keywords in separate ad groups because they're "slightly different", they probably belong together. The algorithm wants concentrated signal.

Negative keywords do more work than your positive keywords

This is the section most posts skip. It's where the budget lives.

Start with the universal negatives. Every account should launch with a starter list. Adjust based on what the business actually sells.

• ✓
  Competitor brand names (unless running competitor campaigns intentionally)
• ✓
  Jobs, career, salary, hiring
• ✓
  Free, cheap, DIY, second-hand
• ✓
  Login, sign in, account, customer service
• Reviews, complaints, scam (if those terms hurt rather than help)
• Your own brand name (if running brand in a separate campaign)
• Anything explicitly off-strategy (e.g. wrong region, wrong product)

Then grow the list weekly. Pull the search terms report. Sort by spend with zero conversions. Add the worst offenders as negatives. Adding 15 to 30 negatives per week is normal for the first 6 to 8 weeks of a campaign. After that the curve flattens.

Pick the right match type for each negative. Negatives have their own match types, and they work differently to positive keywords.

• Negative exactremoves a specific query, word-for-word.
• Negative phraseblocks a 2 to 3 word combination in order.
• Negative broadblocks any query containing all the words in any order.

If you're unsure, default to negative phrase. Negative broad can over-block; negative exact misses obvious variants.

Use shared negative lists. Build one account-level list per concern: jobs, competitors, free-seekers, off-region queries. Apply across campaigns. Performance Max accepts up to 10,000 negative keywords per campaign and shared lists since the 2025 expansion. Use that headroom.

Picking match types in 2026 without bleeding budget

The 2026 match-type playbook:

• Broad match.Only with Smart Bidding and at least 30 monthly conversions feeding the algorithm. Below that threshold broad will spend without learning.
• Phrase match.The safest default. Use when you want some flex but not the full broad firehose.
• Exact match.For transactional head terms you know convert. Smallest list, highest priority, tightest control.
• Modified broad match.Retired in 2021. Don't reach for it.

The Smart Bidding pairing for most accounts is Maximize Conversions with a Target CPA. It's the workhorse. Manual CPC still has a place when you don't yet have enough conversion data to feed an automated bidder, but you should be migrating off it as soon as you have 30 conversions a month.

How keyword research differs for Performance Max

Performance Max is a different animal. You don't pick keywords. You signal intent.

What you actually control inside Performance Max:

• Audience signals.Customer lists, in-market segments, custom segments built from search terms. These are how you point the algorithm at people who look like buyers.
• Asset groups.Themed by intent (demo intent, comparison intent, problem-solving intent), not by keyword.
• Final URLs.The landing pages the algorithm can route to.
• Negatives.Up to 10,000 per campaign plus shared lists.
• The brand-exclusion list.Stops Performance Max cannibalising the brand traffic your Search campaigns already win.

The good news is your keyword research still feeds the work. The same intent buckets (transactional, commercial investigation, informational with buyer signals) tell you which audience signals to build and which asset themes to group around. The output gets fed in as signals rather than as bid targets.

The weekly review loop

The list isn't done when you launch. The weekly review is where good accounts pull ahead of competent ones.

Three things to do every week for the first two months, then bi-weekly after that:

1. Search terms report.What queries are you actually matching? Promote winners into their own ad group. Negate the irrelevant. Add new converters as fresh keywords.
2. Negative additions.15 to 30 a week is normal early on, fewer later.
3. Spend versus conversion check.Any keyword spending more than twice your target CPA without a conversion in 30 days gets paused or moved to a lower bid.

Monthly keyword reviews are now considered table stakes by most experienced PPC practitioners. The reason is that search behaviour shifts faster with AI-driven query patterns, and the search terms report rewrites your keyword list within 30 days whether you read it or not.

The tools we actually use

Five tools, honestly described, beats a 12-tool comparison table no one finishes reading.

• Google Keyword Planner.Free. Has Google's own auction data. Still the floor.
• Ahrefs Keywords Explorer.Paid. We use it for traffic potential, SERP-feature flags, and parent-topic mapping. Strongest for cross-channel research where you want to see the SEO picture alongside the PPC one.
• Semrush Keyword Magic Tool.Paid. Useful when the client already has access. We don't run both Ahrefs and Semrush in parallel.
• Reddit and niche forums.Free. The fastest way to find the vernacular buyers actually type, especially for technical or niche B2B verticals.
• The search terms report.Free, attached to the live account, and underused by 80% of advertisers we audit. Open it every week.

If you only have access to two of those, Google Keyword Planner and the search terms report are the two.

Common mistakes we still see in 2026

Five patterns we still see in inherited accounts every month:

• Building a 500-keyword list when 50 would do.The algorithm wants concentrated data. Long lists fragment your signal.
• Skipping the negative-keyword work because it's "boring".The budget lives in the negatives. Skipping them is skipping the easiest profit you'll find.
• Running broad match without Smart Bidding.The worst of both worlds. Either pair them or use phrase.
• Brand keywords with no protection.Letting competitors bid your brand uncontested is leaving CTR and customers on the table.
• Treating the keyword list as a contract.It's a starting hypothesis. The search terms report will rewrite it within 30 days.

PPC keyword research in 2026 is less about producing a long, beautiful list and more about producing a small, defensible one that the algorithm can act on, plus the negatives that keep it honest. The boring discipline beats the impressive spreadsheet every time.

If you also run organic search, the SEO side of the keyword question is a useful counterweight. We wrote about that separately in our SEO keyword research guide published this week. The two disciplines share vocabulary but reward different habits.

Frequently asked questions