Direct Answer
AI-era research needs two connected maps: a keyword map for repeated Google demand and a prompt map for the conversational situations in which AI systems choose sources and brands. Build the keyword list with seeds and modifiers, vet it with Business potential, Intent, and Difficulty, then apply an AI filter that asks whether the result still earns a click or should be targeted for visibility inside the answer.
The final roadmap should have three queues: click-first SEO opportunities, mention-and-citation AEO opportunities, and hybrid topics that need both. Exact-match prompt chasing is too brittle. Build evidence and content around the full topic, audience, task, constraints, and decision stage.
Watch: Keyword and Prompt Research for AI SEO
Credit: This guide is based on Ahrefs' AEO Course lesson with Sam Oh and the supplied transcript. The three-queue roadmap, prompt matrix, scorecard, and evidence rules below are editorial additions. Start with the preceding brand gap analysis lesson if the brand's missing topics and entities are not yet mapped.
1. Build a Scalable Keyword List
Begin with broad seeds that describe the market, product, service, or customer problem. Combine them with modifiers that reveal the content or action a searcher expects. AI can help brainstorm this vocabulary, but its suggestions are hypotheses until a keyword database, Search Console, customer evidence, or a live result page validates them.
| Modifier family | Examples | Likely need |
|---|---|---|
| Learn | how, why, guide, tutorial, examples | Explanation or instruction |
| Compare | best, top, versus, alternatives, review | Shortlist or decision support |
| Act | calculator, checker, generator, tool, template, finder, planner, maker | Complete a task |
| Buy | price, cost, quote, demo, trial, near me | Commercial or transactional action |
| Fit | for agencies, for beginners, for teams, for a location | Audience-specific suitability |
| Problem | fix, error, not working, improve, reduce, replace | Resolve a specific failure or constraint |
In Ahrefs, enter the seeds in Keywords Explorer, open Matching Terms, and use the Include filter for relevant modifiers. Export the useful candidates with their country, volume, traffic potential, intent, Keyword Difficulty, and current result-page features. Keep the seed and modifier that produced each idea so the research can be repeated.
2. Apply the BID Method
| Test | Decision question | Evidence | Reject when |
|---|---|---|---|
| B: Business potential | If this page wins, can it naturally help the right customer and support an offer? | Product fit, buyer stage, conversion path, first-party revenue or lead data | The traffic is unrelated or the offer must be forced into the answer |
| I: Intent | Can the site create the content type, format, and angle the result page rewards? | Live SERP, dominant page type, format, freshness, local and commercial signals | The required experience conflicts with the planned page |
| D: Difficulty | Can this site produce a credible result and earn the authority needed to compete? | Page-level referring domains, competing brands, content quality, topical depth, unique proof | The gap in authority, product, evidence, or resources is unrealistic |
Ahrefs' formal Business Potential model uses a 0-3 score based on how naturally the product can be presented as the solution. Keyword Difficulty is useful, but it is only a backlink-based estimate in Ahrefs. Read the actual result page. A low score does not make weak content rank, and a high score does not make a valuable topic impossible.
A Transparent Candidate Score
Use the BID decision first. For candidates that pass, rank the work queue with an inspectable score:
Priority = business potential + intent fit + traffic or visibility potential + evidence advantage - production effort - authority gap.
Score each factor from 0 to 5 and retain the factor values. The total is a discussion aid, not an automated publishing decision.
3. Apply the AI Filter
Before approving a keyword, inspect the live result page and ask: can the AI answer fully satisfy this task? Ahrefs' 146-million-SERP study found AI Overviews on 21% of tracked keywords, including 57.9% of question queries and 46.4% of queries with seven or more words. In that dataset, 99.9% of AI Overview keywords were informational.
Those figures describe a dated dataset, not a universal rule. Search features vary by country, device, account, and time. Record the observation date and screenshot or export. Then evaluate the task:
- Answer-complete: a concise definition or fact may produce little reason to click.
- Answer-assisted: the summary helps, but the reader still needs detail, proof, comparison, or implementation.
- Action-required: the reader must use a calculator, checker, generator, template, dataset, product, or service.
- Trust-required: the decision needs firsthand evidence, independent reviews, local expertise, or current professional judgment.
Do not reject every informational topic. Some build authority, earn citations, support a product cluster, or introduce a qualified audience. The filter decides the objective and format, not merely whether a page gets published.
4. Split the Roadmap Into Three Queues
| Queue | Primary outcome | Good candidates | Measure |
|---|---|---|---|
| Click-first SEO | Qualified organic visits and actions | Tools, templates, local services, product pages, calculators, deep comparisons | Rankings, clicks, qualified sessions, conversion |
| Mention/citation AEO | Brand inclusion and source visibility inside answers | Definitions, category questions, “best” and alternatives prompts, original statistics | Mentions, citations, Share of Voice, branded demand |
| Hybrid | Visibility in the answer and a valuable reason to continue | Research, detailed guides, interactive assets, product-led education | Citations plus clicks, assisted conversions, enquiries |
A citation is not a recommendation, and neither guarantees a click. Ahrefs' later self-promotional content experiment found examples where a page was cited while a competitor named on that page was recommended. Report citations, brand mentions, traffic, and conversions as separate stages.
5. Find AI Mention Opportunities
In Brand Radar, add the brand and its domain, choose the relevant AI platform, and isolate responses where competitors appear but the brand does not. Filter commercial-research language such as best, top, versus, review, and alternative. Read the complete response and every cited source before adding an opportunity.
The lesson highlights listicles because Ahrefs reported that “best” lists appeared in 43.8% of ChatGPT source types and 48.9% of AI Overview sources in the cited study. That does not mean any listicle will be trusted or that self-ranking a product first is durable. Prefer independent, maintained sources and accurate inclusion. Where owned comparison content is appropriate, make selection criteria explicit and treat competitors honestly.
Citations are volatile. Ahrefs later reported that only about 54% of cited URLs carried over between consecutive AI Overview checks in a 43,000-keyword study. Save the query, platform, response, cited URL, location, and date. Re-run a fixed set rather than treating one answer as a permanent win.
6. Research Prompt Families, Not Exact Phrases
Conversational systems receive long, contextual requests that may never repeat verbatim. They can also fan one prompt into multiple retrieval queries. Build a matrix of situations around the topic:
| Dimension | Questions to map | Example |
|---|---|---|
| Audience | Who is asking, and what do they already know? | Solo consultant, agency owner, enterprise team |
| Task | What are they trying to understand, compare, choose, or do? | Choose a rank tracker |
| Constraint | What limits the answer? | Budget, geography, privacy, team size, integration |
| Stage | Are they learning, shortlisting, validating, or buying? | First shortlist versus final vendor check |
| Alternative | What other category, workflow, or competitor could solve it? | Spreadsheet versus SEO platform |
| Evidence | What proof would make the answer credible? | Current pricing, tests, case study, independent review |
Create representative prompts from the matrix, then track them consistently. The content plan should cover the shared subtopics and evidence the prompt family requires. Do not create dozens of near-duplicate pages for wording variations with no distinct user value.
The Keyword and Prompt Research Worksheet
| Field | What to record |
|---|---|
| Topic and source | Seed, modifier, customer evidence, competitor page, or brand-gap finding |
| Keyword evidence | Country, volume, traffic potential, intent, KD, SERP features, observation date |
| BID decision | Business score and rationale, intent match, difficulty evidence |
| AI task type | Answer-complete, answer-assisted, action-required, or trust-required |
| Prompt family | Audience, task, constraint, stage, alternatives, and required proof |
| Visibility gap | Missing mention, missing citation, competitor advantage, or absent topic coverage |
| Queue and asset | SEO, AEO, or hybrid; page, tool, video, comparison, dataset, outreach, or update |
| Success measure | Clicks and conversions, mentions and citations, or both, with review date |
A Repeatable Monthly Cycle
- Refresh the seed and modifier list from customer language and product changes.
- Expand and vet candidates with BID using the current live SERP.
- Apply the AI task filter and assign each approved topic to a queue.
- Review Brand Radar mention and citation gaps with the underlying responses.
- Update the prompt-family matrix and re-run the fixed benchmark prompts.
- Ship a small batch, then measure visibility and business outcomes separately.
Continue the Ahrefs AI SEO Course
These requested companion lessons provide the wider strategy around keyword and prompt selection.
1. The New SEO Playbook for AI Search
2. The Google Update That's Changing SEO
3. Give Me 8 Minutes and You'll Win at SEO
Video Chapters
| Time | Lesson | Time | Lesson |
|---|---|---|---|
| 00:00 | Keyword and prompt research | 03:02 | Apply the AI filter |
| 00:33 | Build the keyword list | 04:02 | Find action-first queries |
| 01:37 | Vet candidates with BID | 05:02 | Find AI mention opportunities |
| 01:57 | Business potential | 06:27 | Citation volatility |
| 02:23 | Intent and difficulty | 06:45 | Prompt research and fan-out |
Sources and Further Reading
- Ahrefs: Keyword and Prompt Research for AI SEO (primary video and supplied transcript)
- Ahrefs: Keyword Research, the Beginner's Guide
- Ahrefs: Keyword Intent; Keyword Difficulty
- Ahrefs: How to Rank in AI Overviews
- Ahrefs: 2026 SEO Trends and Listicle Citation Data
- Ahrefs: Self-Promotional Content AI SEO Experiment
- Ahrefs Help Center: Brand Radar
YouTube lists the primary video's publication date as 6 May 2026. This article and its data notes were reviewed on 27 September 2026. Search results, AI answers, citations, product interfaces, and datasets change; retain dates and exact test conditions with every research export.