Most keyword research advice on page one of Google is still 2018 thinking. Type a few seeds into a tool, sort by volume, filter by difficulty, write content. That workflow broke the moment AI Overviews started eating the click-through rate, AI Mode started answering queries without surfacing pages, and the SERP filled up with Reddit threads instead of editorial guides.
If you've opened a keyword tool recently and felt overwhelmed, you're not the only one. The workflow most guides describe was last fully useful around 2020.
This is the playbook we actually use when we onboard an HJD client today. What we check before search volume. How we use the paid tools (and where they're a trap). The five-question filter we run before recommending any content. What's changed because of AI search.
Most keyword research advice predates AI search
The canonical workflow looks like this: open a keyword tool, find a head term with high volume, check the difficulty score, pick something winnable, write a 3,000-word guide. That sequence assumes the reader will type a query, see your blue link in the top three, click through, and convert.
A meaningful share of queries don't work that way anymore. The position-1 result for "keyword research for seo" right now is an AI Overview. The reader gets their answer without clicking anything. Two of the top 10 results are Reddit threads. Half the rest are tool product pages from Ahrefs and Semrush rather than editorial guides.
The old workflow isn't dead. It's just become the last step now, not the first.
The new mental model
Keyword research in 2026 isn't "find high-volume keywords." It's finding the entities, intents, and questions Google associates with your domain, and deciding which of them you can credibly own. The keyword tool comes last.
What we check before search volume
Before we open Ahrefs we run four checks. None of them are about keywords.
Who Google thinks we are
Google indexes your domain as a collection of entities, not a list of keywords. If your site is mostly about plumbing, Google sees you as a plumbing entity. Trying to rank a great post about keyword research from a plumbing domain doesn't work, no matter how good the post is, because the entity authority isn't there.
How we check: pull the top 50 to 100 keywords the domain currently ranks for and cluster them by topic. That cluster is the entity profile. If you want to win for a query and the entity behind it sits two clusters away from your current authority, the keyword work is the wrong place to start. Fix the entity gap first.
Tools we use here: Ahrefs site-explorer-organic-keywords if budget allows, otherwise Google Search Console's Performance and Queries report (free, mandatory anyway).
What answer the reader actually wants
The query is the question form. The intent is the answer the reader actually needs. In 2026 those often diverge.
"Keyword research for seo" looks like a textbook informational query. Open the SERP and half the results are tool product pages from Ahrefs and Semrush. The reader isn't just looking for an explanation. They're partly evaluating which tool to use. A pure explainer with no opinion on tools misses half the SERP intent.
We do a manual SERP scan for the top 5 candidate keywords. Each result in the top 10 gets a classification: editorial guide, tool page, community thread, tool listicle, or case study. The underserved slot tells us where the angle lives. For this post, that slot was agency-practitioner editorial with an opinion. There isn't a credible one in the current top 10.
Whether AI Overviews are eating the SERP
If position 1 or 2 is an AI Overview, the click-share maths falls apart. The reader sees a summary at the top of the page with three to seven cited sources. They get the answer there. A click to your page becomes a marginal event.
The play changes. Instead of ranking to win the click, the goal becomes being a cited source in the AI Overview. That's a different content shape: extractable answers up top, depth underneath.
How we check: search the candidate keyword logged out, in incognito, in the target country. Note whether an AI Overview appears, which sources it cites, and whether your site could plausibly be one of those sources within six months. If the cited sources are all DR 90+ household names and your site is DR 25, this query isn't going to pay back even with a great post.
Whether Reddit owns the SERP
Two of the top 10 results for "keyword research for seo" are Reddit threads. That's a signal from Google. For this query, the reader wants real-person perspective, not a polished agency guide. Either we change the angle to feel more like a real practitioner talking, or we skip the keyword.
We sometimes seed our own answers in relevant subreddits when this happens. Not as link bait, and not in disguise. As part of doing the keyword research itself. If the Reddit thread for our target query has 40 comments, that's 40 real audience-listening data points. We read them.
How we actually use the keyword tools
We use Ahrefs every day. The tools aren't the problem. The order is. We use them to verify a hypothesis we already have, not to find a topic from a cold start.
The workflow looks like this. Seed terms come from somewhere real: sales calls, support tickets, the GSC Queries report. We expand those seeds in Ahrefs keywords-explorer-matching-terms and filter on KD and parent topic. Not on volume. We cluster the result by entity, pick one cluster the client can credibly own, then pick one specific mid-tail term inside it.
The volume-first approach is how a 10-person team ends up trying to rank for "keyword research" (Vol 604,000, KD 92) and writing the 31st generic guide that nobody needs. KD 92 means the current top 10 are sites doing 70,000+ monthly traffic each. The maths doesn't work for almost any small or mid-sized business.
The mid-tail equivalents are smaller but winnable. "How to do keyword research" is Vol 2,000, KD 69. "Keyword research for seo" (our target with this post) is Vol 2,700, KD 85. These are still hard. Unlike the head term, they have a path.
The five-question filter we run before recommending content
Before we recommend writing anything for a keyword, we answer five questions. If two or more come back weak, we change the angle or skip the topic.
- Does the SERP intent match what we'd actually be writing?Open the top 10, classify each result. If our angle doesn't fit a slot, the post will rank for nothing.
- Is there an AI Overview on this keyword?If yes, we shift the content shape: extractable answers up top, depth underneath. We play for citation, not just clicks.
- Are we credible on this entity?Check current GSC rankings for adjacent terms. A site that ranks 0 to 50 on the topic won't suddenly take a KD 80 head term.
- Can we serve the reader better than the top 3?If the top three are all tool pages and we don't sell a tool, the reader needs editorial. If they're all generic editorial, we win on specificity and POV.
- What's the payoff if we don't rank but still publish?Sometimes a post is worth publishing at KD 80 because it supports the entity profile, earns links, or gets cited in AI Overviews. Volume isn't the only payoff.
This filter does most of the work. The keyword tool tells us volume and KD. The filter tells us whether the post is worth the next two weeks of editorial work.
What AI search has changed
From keywords to questions
AI search runs on questions, not keywords. Pull the People Also Ask block for the head term and you'll see the actual phrasing readers use. Use those phrasings verbatim as H2 candidates.
That sounds like a small thing. It changes the structure of the post. The H2s become answers to specific questions rather than generic section headings, and AI models can lift them as discrete answer units. "What does keyword research mean?" is a worse H2 than "How keyword research works in 2026." The second has a clear answer attached.
From volume to citation
AI Overviews cite three to seven sources per query. Those citation slots replace some share of the click slots for a meaningful fraction of search traffic. The work shifts from winning the click to being one of the three cited sources.
How: write a 30 to 50-word extractable answer under each H2. Direct, specific, no hedging. Something a model can lift verbatim without rewriting. The depth still goes underneath for readers who want it. The extractable answer at the top is for the model.
From long-form to short answer plus depth
The old play was 3,000-word exhaustive guides because Google's ranking signal partly rewarded comprehensiveness. The new play is the short extractable answer up top, then depth for the reader who scrolls.
Shorter posts aren't the goal. Structurally different posts are. A 2,000-word post that answers a question in the first 50 words and then earns the next 1,950 with depth beats a 3,500-word post that buries the answer in section four.
A 30-minute version of all of this
If you only have 30 minutes today and you want to do keyword research the right way for one keyword, here it is.
Pull Performance and Queries. Note the top 30 queries the site already shows up for, with impressions and click rates.
Group queries by topic. Two or three clusters is normal. Note which cluster has the most impressions but the fewest clicks.
Pick a mid-tail keyword inside it that the site doesn't yet rank for. Avoid head terms. Pick something specific.
Note whether an AI Overview appears, what sources it cites, the intent split in the top 10, and whether Reddit is showing.
Use the five-question filter. If the keyword passes, write the post with extractable answers up top and depth underneath. If it doesn't, skip it and try the next mid-tail.
That's the lazy version. It won't replace a proper keyword strategy. It will tell you whether the topic you're about to write is worth the time.
The tools we use, and where we'd save the money
We use the paid tools. Most agencies do. Honest assessment of when the money is worth it:
| Free | Paid (single tool) | Paid (agency stack) | |
|---|---|---|---|
| What you get | GSC + Ahrefs Webmaster Tools + AnswerThePublic | Ahrefs or Semrush (one of them) | Ahrefs + Brand Radar + a SERP tracker |
| Who it fits | Solopreneurs, pre-revenue businesses, side projects | In-house marketing teams with one site to grow | Agencies, brands with multiple properties |
| Where it falls short | No competitor keyword data, limited keyword expansion | Single-domain view, no cross-client comparison | Cost stack adds up; only worth it past a certain client count |
A quick note. When we name Ahrefs or Semrush we name them honestly. Both are excellent. We use Ahrefs because we've used it longer, not because we've measured it as better on any dimension that matters. If you're starting from zero, pick one based on the trial UX and the price you can stomach. Don't run both. Don't subscribe to either before you've spent a month with the free tools first.
For AI Overview tracking specifically, the tooling is still maturing. Ahrefs Brand Radar is the cleanest option we've tried. A manual log in a Google Sheet still works fine for one or two priority queries.
What this looks like with a real client
A regional B2B services firm came to us last quarter with what they called a content backlog. It was a list of topics their junior had written down from a webinar. One was "best practices for keyword research."
We killed it before drafting. The firm's entity profile was zero-percent SEO-adjacent. Their domain ranks for terms in their core service categories, and that's it. Even a great post on keyword research would have ranked nowhere and earned no AI Overview citations. The post would have cost two weeks of editorial time and returned nothing.
We picked three entity-fitting clusters straight from their GSC instead, one per core service category. We wrote one mid-tail post per cluster. Two of the three earned top-5 rankings inside four months. The third is still climbing. The keyword research work, in this case, was the filter. The product was knowing what not to write.
The reset
The starting point for keyword research isn't a keyword tool. It's a five-minute audit of who Google thinks you already are.
Entities first. Reader intent next. AI Overview check after that. Keyword tools last, and only to verify what you've already decided.
The tools haven't got worse. The work around them has changed.
Want a second pair of eyes on your keyword strategy?
We'll audit your current rankings, your entity profile, and the AI Overview exposure on your top queries. Free first conversation.
Book a chat
