Most small business owners approach keyword research the same way they did five years ago. They type something into a tool, look for high volume and low difficulty, and start writing. That worked once. In 2026, it’s a half-finished strategy.
The reason is simple: there are now two search engines you need to win. Google, where people click links. And AI platforms like ChatGPT, Perplexity, and Google’s AI Mode, where people get answers directly without clicking anything. Getting visible in both requires a different research process, and the good news is that the two overlap more than you’d think.
This article walks you through the full keyword and prompt research process Fuzelift uses with clients, built on an Ahrefs-developed framework and sharpened with fresh data on how AI engines actually decide which pages to cite.
Key Takeaways
- You now have two search engines to win. Google ranks you. AI engines cite you. A strategy that ignores GEO is already leaving half your potential audience on the table.
- Every keyword must pass the B-I-D filter. Business potential, Intent match, and Difficulty. Fail any one of them and the keyword isn’t worth your time, regardless of search volume.
- The AI filter is the step most guides skip. If AI fully answers a query in its own response, there’s no click to compete for. Tool-based queries (calculators, checkers, templates) are the exception because AI can describe a tool but can’t let you use one.
- Getting cited by ChatGPT follows clear rules. 88% of citations come from the standard search index. Pages that match the AI’s internal sub-questions get cited. Pages that don’t get retrieved and ignored. Your title and URL structure matter more than most people realise.
- Topic authority beats individual pages every time. AI builds consensus from multiple sources. One good article won’t cut it. A bunch of interlinked content around your core topic is what earns you consistent visibility in AI responses.
- Prompt research is not keyword research. People talk to AI in full sentences with context. You can’t optimise for a single prompt. You build authority across a topic so that however the question is phrased, your brand appears in the answer.
Why Keyword Research Has Changed in 2026
The old game was: find a keyword, write an article, wait for traffic. That model hasn’t disappeared, but it’s been complicated by one significant shift.
AI search engines, including ChatGPT, Google AI Overviews, and Perplexity, now intercept a growing percentage of searches. According to Ahrefs data, AI overviews appear on around 21% of all keywords, but that number jumps to nearly 58% for question-based queries and 46% for searches with seven or more words. For most informational searches, AI is the first answer the user sees.
If your business isn’t being cited in those AI answers, you’re invisible for half of your potential audience before they’ve even had a chance to find you.
The solution isn’t to abandon traditional SEO. It’s to add a second layer: Generative Engine Optimisation, or GEO, which is the practice of making your content citable by AI systems, not just rankable by Google.
Step 1: Build Your Keyword List the Right Way
The starting point hasn’t changed much. You need two things: seed keywords and modifiers.
Seed keywords are the broad terms at the heart of your business. If you’re a plumber in Manchester, “plumber Manchester” and “emergency boiler repair” are seeds. If you run an e-commerce store selling running gear, your seeds might be “running shoes,” “trail running kit,” and “marathon training.”
Modifiers are the words that transform a broad seed into a specific, useful search query. Words like “best,” “how to,” “near me,” “cost of,” “vs,” “review,” and “calculator” pull your seeds toward real searches with real intent behind them.
A fast way to generate both: ask an AI assistant to give you ten seed keywords for your type of business and five or more modifiers relevant to your niche. You’ll have a solid working list in minutes.
From there, paste those seeds into a keyword research tool, run them through the matching terms report, and apply your modifiers as filters. You’ll go from a handful of ideas to hundreds of real keyword opportunities your potential customers are already searching for.
Step 2: Vet Every Keyword with the B-I-D Formula
Not all keywords are worth chasing. Some look great on paper and waste your time in practice. Before committing to any keyword, it needs to pass three tests. Think of this as the B-I-D formula.
B is for Business Potential. Ask yourself: if I rank number one for this keyword, does it actually help my business? A search like “what is a conversion rate” might have decent volume, but someone asking that question is probably a student or a curious marketer, not a small business owner ready to hire an agency. Compare that to “Google Ads conversion rate by industry” where the searcher clearly has a campaign running and wants to benchmark it. Always chase keywords where the person searching could realistically become a customer.
I is for Intent. Google the keyword and look at what’s actually ranking. If you see product pages and every result is from Amazon or an e-commerce store, Google has decided that query has transactional intent. Writing a blog post won’t rank there, regardless of how good it is. The results page tells you what format Google wants. Match the intent or move on.
D is for Difficulty. Look at the domain ratings and referring domains of the pages already ranking. If the top ten results all belong to major media outlets with thousands of backlinks, you’re looking at a multi-year project, not a quick win. A keyword where a few lower-authority sites are ranking in the top ten is a sign you can compete.
If a keyword passes all three, it’s worth targeting. But there’s one more filter to apply before you commit.
Step 3: Apply the AI Filter Before You Write a Single Word
Here’s the question most keyword guides skip: can AI fully satisfy this search without the user ever clicking through to your site?
If someone searches “what is a conversion rate” and ChatGPT or Google’s AI Overview gives them a complete, accurate answer in three sentences, there’s no click to compete for. The keyword passes B-I-D but fails the AI filter. Writing a 2,000-word article targeting that query is work you’ll do for almost zero return.
Before committing to any keyword, search it yourself. Look at what shows up. If the AI answer is complete, the keyword is a trap.
The exception: tool-based queries. Searches like “Google Ads budget calculator,” “keyword difficulty checker,” or “lead gen conversion calculator” still generate clicks because AI can describe a tool, but it can’t let you use one. That’s why the Fuzelift Google Ads Calculator attracts traffic that pure informational articles on the same topics can’t. When someone searches for a calculator, they need to open something and interact with it. AI can’t replace that (yet).
To find these opportunities in your niche, add modifiers like “calculator,” “tool,” “template,” “generator,” “checker,” and “planner” to your keyword research. These are low-competition, high-intent queries that AI consistently leaves open for organic results.
Step 4: Find Your AI Mention Gaps
Passing the AI filter doesn’t mean ignoring AI-heavy queries. It means targeting them differently.
For queries where AI overviews dominate, your goal shifts from “rank and get a click” to “get cited inside the AI answer itself.” That’s GEO. And the first step is figuring out where your competitors are showing up in AI responses and you aren’t.
Tools like Ahrefs Brand Radar let you see exactly which queries are generating AI mentions for your competitors. Filter for queries containing “best,” “top,” “vs,” “review,” or “alternative,” and you’ll see the shortlist of searches where AI is recommending other businesses in your space while ignoring yours.
That’s your content priority list. Not the keywords that look good in a spreadsheet. The queries where AI is already making recommendations and you’re not part of the conversation.
Step 5: Understand How AI Actually Chooses Which Pages to Cite
This is where most guides go quiet, because the answer requires more than guesswork. Fortunately, a large-scale Ahrefs study of 1.4 million ChatGPT prompts gives us real data to work with – but I’ll make this 3300 word article more straightforward for you.
The study found that ChatGPT only cites about 50% of the URLs it actually retrieves. It pulls in dozens of pages per query, reads some, and cites roughly half. The question is which half.
The research identified a few clear patterns.
Search ranking still matters. 88% of all URLs cited by ChatGPT come from its general search index. In plain terms: if you don’t rank, you don’t get cited. The AI isn’t going rogue and finding obscure pages. It’s mostly working from the same pool of content that ranks in Google search.
Title relevance to sub-questions is critical. When you type a prompt into ChatGPT, the AI internally generates multiple sub-questions, sometimes called “fanout queries,” to answer it fully. A question like “what’s a good Google Ads budget for a small business?” might generate sub-questions about average CPCs, typical ROAS benchmarks, budget split recommendations, and industry-specific considerations. The study found that cited pages had consistently higher semantic similarity between their title and these fanout queries than non-cited pages. The gap between cited and non-cited pages was even wider for fanout queries than for the original prompt.
What this means practically: your page title and URL need to match not just the broad topic, but the specific sub-questions someone might ask within that topic. “Google Ads budget calculator for small business” is a more citable title than “Google Ads budgeting guide” because it maps to a more specific sub-question.
Human-readable URLs help. The study found that pages with natural language slugs (think /google-ads-budget-calculator/ rather than /page?id=4732) had an 89.78% citation rate compared to 81.11% for opaque URLs. A gap of nearly nine percentage points from something as simple as your URL structure.
Freshness matters, but only up to a point. ChatGPT does skew toward newer content across its full citation pool. But within a single prompt’s retrieval set, it tends to cite more established pages over very new ones. Relevance does more work than recency. A new page that precisely matches the fanout queries will get cited. A new page that doesn’t match will be retrieved and then ignored.
Step 6: Build Topic Authority, Not Just Individual Pages
Here’s the part most small businesses miss entirely, and it’s the reason they write a dozen articles and still can’t get AI to mention them.
AI citation isn’t just about the quality of one page. It’s about whether your site is recognised as an authoritative source on a topic. AI systems are trained to build consensus. When ChatGPT sees your brand mentioned across multiple list-based articles, comparison posts, and industry roundups, it starts to register you as a credible reference, not just a page it stumbled across.
The Ahrefs research found that 43.8% of all pages cited by AI are listicles. “Best Google Ads agencies for small businesses.” “Top SEO services for e-commerce.” “Best tools for tracking PPC campaigns.” These pages appear so often because they represent consensus. Multiple sources recommending the same brand is a signal AI picks up on and repeats.
Building topic authority means “owning” a bunch of content around your core subject, not just having one article. It means being mentioned on third-party sites (according to multiple studies, think of YouTube, Reddit, LinkedIn, and more). It also means your Google Business Profile is complete and consistent (because local citations feed into AI’s understanding of who you are and where you operate). And it means your content genuinely covers the sub-questions AI is asking, not just the surface-level keyword.
For local businesses especially, Google Business Profile optimisation plays a bigger role in AI visibility than most people realise. When someone asks ChatGPT or Google AI Mode “who is the best plumber near me,” the answer is partly shaped by how well-maintained and consistent your GBP listing is. That’s not speculation. It’s a direct consequence of how AI systems verify local business credibility.
Step 7: Research Prompts, Not Just Keywords
The final layer of this process is prompt research, and it’s genuinely different from keyword research.
When someone opens ChatGPT or Perplexity, they don’t type “best SEO agency.” They ask something like “I’m a small business owner in Dublin, I’ve been running Google Ads for two years but I’m not sure if I’m getting good results, what should I actually be measuring?” That’s a prompt. It’s natural, contextual, and far more specific than any keyword.
The same question asked ten different ways can produce ten different AI answers with ten different brands mentioned. You can’t optimise for a single prompt the way you optimise for a single keyword.
What you can do is build visibility across an entire topic so that however the question is phrased, your brand has a claim to being part of the answer. That means having content that answers the question at every level of specificity: the broad overview, the comparison angle, the calculator or tool, the specific “how to” guide, and the case study that shows you’ve done it in practice.
This is exactly why Fuzelift’s approach to SEO and AI Search Optimisation treats topic clusters as the unit of work, not individual pages. A single article targeting a single keyword is a bet. A cluster of interlinked content built around a core topic is a moat.
What This Means for Your Small Business Right Now
If you’re running a small business and you’re serious about being found in 2026, here’s the honest summary of where to focus:
The keywords worth targeting are the ones that pass B-I-D, survive the AI filter, and connect to a topic where you can genuinely build authority. The prompts worth targeting are the ones your ideal customers are already having with AI systems, especially the comparison and recommendation queries where your competitors are showing up and you aren’t.
Getting this right requires a systematic process, the right tools, and ongoing attention. AI citations refresh on average every two days. Fanout queries change. Competitor content shifts. This isn’t a set-it-and-forget-it exercise.
If you want a team that actively manages this process for you, Fuzelift’s SEO and AI Search Optimisation service handles the full keyword and prompt research process, builds the content cluster, and monitors your AI visibility alongside your organic rankings. For local businesses, our Google Business Profile management service ensures the local signals AI systems use to verify your credibility are always accurate, complete, and working in your favour.
The businesses showing up in AI answers in 2026 aren’t doing anything mysterious. They’ve built real authority around a real topic, with content that precisely maps to the questions AI is asking behind the scenes. That’s the standard. Everything else is just hoping you get lucky.
Not sure where your business stands in AI search? Book a free 20-minute call with the Fuzelift team. We’ll look at your current visibility, understand your market and goals, and tell you honestly whether we can help.