How to Do Prompt Research for AI SEO

Prompt investigation is the process of identifying and search the questions that origin AI systems to comparison options and urge circumstantial brands. 

The punctual investigation process serves the aforesaid foundational domiciled for AI visibility that keyword investigation serves for SEO and PPC but the portion of measurement is different. Instead of pages and queries, punctual investigation focuses connected really AI systems shape and coming recommendations.

In AI SEO, visibility only matters erstwhile AI is evaluating choices. That’s erstwhile it weighs alternatives, applies constraints, and points personification toward a solution. If your marque isn’t coming successful those moments, it won’t facet into the decision.

Most prompts ne'er scope that stage. They make explanations, summaries, aliases wide guidance. Prompt investigation filters those retired and focuses connected middle- and bottom of the chimney (BOFU) prompts: comparisons, evaluations, and "best" queries wherever AI weighs alternatives and recommends solutions.

To show really punctual investigation useful successful practice, I’ll locomotion done the nonstop process I usage to way Semrush’s ain LLM visibility growth. 

My punctual investigation process follows 4 steps:

• Identify target audiences and purchaser personas
• Describe solutions and really they thief those audiences
• Use keyword investigation arsenic supportive connection input
• Use an LLM to make BOFU prompts for tracking

By the extremity of this guide, you'll person a repeatable measurement to build a punctual group that shows wherever your marque competes and wherever it doesn't. But first, let's explain really punctual investigation differs from the keyword investigation you already know.

How Prompt Research Differs from Keyword Research

For hunt marketers, punctual investigation introduces a acquainted conception pinch caller challenges. Unlike accepted search, we don't person years of humanities hunt volume, CPC, aliases inclination information for AI prompts.

Because of that, punctual search doesn’t behave for illustration keyword tracking.

SEO rankings thin to beryllium comparatively predictable. AI-generated answers are volatile and personalized. Prompt investigation focuses connected guidance and shape recognition, not fixed positions aliases precise counts.

The opposition becomes clearer erstwhile you look astatine really the 2 approaches disagree successful practice.

Even pinch these differences, the nonsubjective hasn’t changed. You’re still defining a group of target questions, improving your visibility astir them, and measuring capacity complete time.

What has changed is really visibility is discovered and evaluated.

Semrush has built a prompt database informed by existent clickstream data from ChatGPT and different AI platforms, allowing you to estimate taxable measurement arsenic it happens connected LLMs.

Is Keyword Research Still Relevant?

Yes.

Keyword investigation still plays an important supporting domiciled because it reveals really group picture problems and what intent sits down their searches. 

Those signals thief you determine which prompts are worthy targeting. The quality is that keywords are nary longer the endpoint; they’re a connection input that gets rewritten into natural, conversational prompts.

The larger displacement is what you optimize for.

Instead of search “wins” the measurement you would successful SEO, punctual investigation looks astatine which topics, constraints, and personas consistently urge your brand, and wherever it fails to appear. That’s why punctual investigation prioritizes the ideal customer profile (ICP), aliases type of customer a merchandise is built for, complete cost-per-click. 

The guiding mobility changes from which position are cheapest aliases highest measurement to whether your marque appears for the types of intent that bespeak existent buying situations.

Tracking AI responses complete clip makes that visibility observable. Daily snapshots of AI answers create a moving grounds of really your marque is framed, compared, aliases omitted crossed decision-oriented prompts.

With that instauration successful place, the adjacent measurement is building a punctual group that reflects really your buyers really make decisions.

1. Identify Your Target Audience

Personas specify what questions get asked. That’s existent for keyword investigation and punctual investigation alike. But for punctual research, personas besides find whether AI recommends thing astatine all. 

That’s because constraints are what push AI systems retired of mentation mode and into proposal mode. A generic mobility for illustration “what’s bully canine food?” produces education. A constrained mobility for illustration “best limited-ingredient canine nutrient for a canine pinch tummy issues nether $60/month” forces a comparison. 

Before generating prompts for LLM tracking, attraction connected the persona traits that alteration really AI evaluates options:

• Context & acquisition level: who’s asking and successful what situation
• Primary consequence aliases pressure: what they’re trying to debar aliases resolve
• Language & expertise: casual vs. method phrasing
• Budget expectations: affordable, mid-range, aliases premium

For example, a canine genitor managing nutrient sensitivities mightiness input, "best limited-ingredient canine nutrient for tummy issues." A different proprietor feeding patient ample dogs whitethorn hunt for "affordable canine nutrient for ample breeds," while a premium shopper focused connected nutrition looks for "human-grade, single-protein caller canine food."

The class stays the same, but the constraints and the recommendations AI returns alteration pinch the persona.

Where to Tap into Persona Characteristics

Buyers uncover really they think, speak, and make decisions successful open, unfiltered spaces for illustration connection boards, reviews, and support discussions wherever they talk astir products successful their ain words.

Personas that consistently uncover consequence management, trade-offs, and uncertainty simplification create the strongest instauration for punctual research. Their constraints people unit AI systems to comparison options and make recommendations.

2. Connect Your Product’s Solutions to Your Persona’s Problems

When group inquire AI to thief them take betwixt options, they’re seldom comparing characteristic lists. They’re trying to determine whether a merchandise fits their situation, reduces risk, and feels for illustration a safe choice.

AI recommendations thin to bespeak that behavior. Brands are suggested much often erstwhile their products intelligibly resoluteness the circumstantial hesitation a purchaser feels astatine the infinitesimal of decision.

Your merchandise needs to beryllium described crossed the sources AI systems trust connected successful position that thief a purchaser decide, not conscionable understand.

 These specifications include:

• Features: What the merchandise delivers successful actual terms.
  These are actual attributes AI tin reference straight (e.g., “single-protein formulas,” “SOC 2 compliant,” “native Shopify integration”).
• Benefits: Why those features matter to the persona.
  Benefits construe features into outcomes that trim interest (e.g., “easier digestion,” “faster onboarding,” “lower implementation risk”).
• Use cases: Situations wherever the merchandise fits cleanly.
  These thief AI lucifer solutions to scenarios (e.g., “for delicate stomachs,” “for mini teams,” “for regulated industries”).
• Problems resolved: The circumstantial risk, friction, aliases uncertainty the merchandise removes.
  This is often the strongest proposal trigger (e.g., avoiding allergic reactions, preventing costly mistakes, reducing vendor lock-in).
• Fit factors: Indications that make the action consciousness safer aliases smarter than alternatives, specified arsenic clarity, simplicity, consistency, aliases alignment pinch purchaser expectations.

Together, these elements picture overmuch of the logic that AI systems usage erstwhile comparing brands.

Validating Which Attributes Matter successful AI Comparisons

If you request thief determining which attributes are driving persona preferences, leverage Brand Performance successful the Semrush AI Toolkit. This instrumentality shows which features AI already emphasizes erstwhile comparing brands successful your category.

For example, for the business Dover Saddlery, AI consistently explains its recommendations utilizing operational fresh indications, for illustration “one-stop assortment depth” erstwhile buyers request aggregate items astatine erstwhile and “fast, reliable fulfillment.”

These are the reasons AI gives erstwhile justifying why Dover is simply a viable prime successful a circumstantial determination context. Collectively, they position the marque arsenic a dependable, low-risk outfitter which is the awesome AI needs to urge a retailer erstwhile the buyer’s privilege is reliability complete exploration.

These attributes go the building blocks for punctual generation. When you provender persona constraints and merchandise fresh factors into an LLM, you springiness it the discourse it needs to make decision-stage prompts, not generic questions.

3. Use Keyword Research to Support Prompt Discovery

Keyword investigation validates connection for punctual investigation by confirming really your assemblage people frames problems alternatively than estimating demand.

Tools for illustration Semrush’s Keyword Magic Tool uncover patterns successful language, including:

• Which constraints look repeatedly
• Which modifiers consciousness earthy versus technical
• Which brand-plus-ingredient combinations show up consistently

Start pinch a taxable tied to a constraint. In this case, “dog nutrient ingredients” reflects really ingredient-sensitive buyers mightiness framework the problem.

Phrases for illustration “limited constituent canine food,” “best constricted constituent canine food,” and “limited constituent canine nutrient for allergies” recur crossed commercialized and mixed-intent searches.

This consistency indicates really buyers successful this niche building their options and modifiers. Keyword investigation helps validate language, but it doesn’t show really AI systems respond to that connection successful practice.

Use Prompt Research to See How AI Responds

Once you’ve identified persona language, participate that wording into the Prompt Research instrumentality to research really AI systems are responding to the topic.

For example, we entered “limited constituent canine food” successful the Prompt Research tool.

In the “Topics” view, AI clusters the class astir formulations and brands, including hypoallergenic diets, constricted constituent products, and brand-specific variants. That building indicates the “limited ingredient” taxable already immunodeficiency decisions, making it a beardown campaigner for a BOFU prompt.

4. Generate a List of BOFU Prompts for LLM Tracking

The Prompt Research instrumentality tin uncover early punctual candidates for a speedy commencement to query selection. Many of these, however, bespeak exploratory questions that don’t make for reliable tracking. Prompts specified arsenic “What should I provender my dog?” seldom correspond a existent determination moment.

Instead, prioritize prompts that present constraints and unit a prime for a circumstantial persona, for illustration “What’s a bully canine nutrient for a canine pinch digestive issues that isn’t expensive?” These are the prompts wherever marque mentions appear, and preferences commencement to form.

Once you tin admit what a trackable punctual looks like, you tin usage an LLM to efficiently make and grow a focused punctual group astatine scale.

How to Generate Decision-Stage Prompts pinch an LLM

Effective BOFU prompts require context. The LLM needs clarity on:

• Who is asking
• What result they’re trying to avoid
• What constraints style the decision
• How the purchaser people describes the problem
• That the mobility must consequence successful a proposal aliases comparison

With that discourse successful place, the output shifts distant from acquisition and toward evaluation.

A champion believe is to usage a accordant pre-prompt to support outputs focused connected BOFU intent. 

For example:

Act arsenic a purchaser investigation assistant. Generate decision-stage questions that would origin an AI strategy to comparison and urge circumstantial brands.

Buyer context:
– Persona: [describe the purchaser and situation]
– Primary consequence aliases concern: [what they want to avoid]
– Constraints: [budget, requirements, exclusions]
– Language cues: [phrases the purchaser uses]

Instructions:
– Do not see marque names successful the questions
– Each mobility must require a proposal aliases comparison
– Avoid acquisition aliases definitional phrasing
– Write prompts precisely arsenic a existent purchaser would inquire them

This template keeps each procreation tally aligned pinch decision-stage output.

If the output still feels acquisition (and not recommending immoderate brand), tighten the constraints and effort again until the exemplary makes a recommendation.

When marque mentions look consistently, and the questions bespeak a existent prime being made, you’ve reached a punctual worthy tracking.

Account for Query Fan-Out successful Your Prompt Set

Query fan-out is the process of really AI systems break a punctual into respective smaller queries, find answers to each, and harvester them into 1 complete response.

When personification asks "best limited-ingredient canine nutrient for allergies," AI systems for illustration ChatGPT and Google AI Mode break that mobility into aggregate sub-queries, which could be:

• Hypoallergenic canine nutrient recommendations
• Single-protein canine nutrient brands
• Grain-free canine nutrient for delicate stomachs
• Dog foods without communal allergens

The AI past retrieves accusation for each sub-query and merges it into a azygous response. This process allows AI to supply richer, much circumstantial answers, moreover erstwhile nary azygous root straight addresses the original query.

Track these variations to spot really good you look for each queries related to intents. If your marque appears crossed variations, you'll person a amended chance of being recommended.

This attack mirrors really AI systems really process queries, helping you build a punctual group that captures the afloat scope of sub-queries AI mightiness make erstwhile evaluating your category.

How Many Prompts Should You Track? 

To understand your AI visibility, way arsenic galore chopped decision-stage prompts arsenic your allowance supports, focusing connected different determination contexts alternatively than insignificant wording variations.

Each Semrush One scheme includes a fixed allowance of tracked prompts. This allowance determines really galore unsocial prompts you tin show astatine the aforesaid clip (for example, 50, 100, aliases 200).

With a smaller punctual allowance, attraction connected prompts that mightiness urge your products aliases services that thrust revenue. 

Track a tight group of decision-focused prompts for each merchandise aliases service. Based connected our soul testing, 10 well-chosen prompts per merchandise are usually capable to spot whether AI systems consistently urge your marque aliases default to competitors.

With a larger allowance, adhd prompts only wherever information criteria change, for illustration persona, industry, aliases usage lawsuit alternatively than utilizing mini wording variations that usually nutrient the aforesaid AI behaviour and don’t create caller signals. 

You tin besides align immoderate tracked prompts pinch keywords you already show successful SEO to comparison hunt visibility pinch AI visibility.

5. Track Your Prompts and Measure Visibility Over Time

Once you've built your punctual set, the last measurement is to set up your LLM punctual tracking to spot really AI responds complete time. 

Semrush offers Prompt Tracking via the Position Tracking tool. 

Start a caller run by entering your target AI level punctual database to track.

Once your run is running, Semrush checks these prompts regular and records whether your marque appears successful the AI-generated response. You’ll spot AI Visibility, Mentions, and Average Position from the Landscape tab.

This helps you measurement wherever you’re present, wherever competitors are winning, and wherever you’re missing visibility.

To study connected progress, it's besides easy to make a PDF from your search campaign. 

How Many Prompts Should You Track? 

To understand your AI visibility, way arsenic galore chopped decision-stage prompts arsenic your allowance supports, focusing connected different determination contexts alternatively than insignificant wording variations.

Each Semrush One scheme includes a fixed allowance of tracked prompts. This allowance determines really galore unsocial prompts you tin show astatine the aforesaid clip (for example, 50, 100, aliases 200).

With a smaller punctual allowance, attraction connected prompts that mightiness urge your products aliases services that thrust revenue. 

Track a tight group of decision-focused prompts for each merchandise aliases service. Based connected our soul testing, 10 well-chosen prompts per merchandise are usually capable to spot whether AI systems consistently urge your marque aliases default to competitors.

With a larger allowance, adhd prompts only wherever information criteria change, for illustration persona, industry, aliases usage lawsuit alternatively than utilizing mini wording variations that usually nutrient the aforesaid AI behaviour and don’t create caller signals. 

You tin besides align immoderate tracked prompts pinch keywords you already show successful SEO to comparison hunt visibility pinch AI visibility.

Turn the Growth of AI Into an Actionable Signal for Your Marketing

As AI platforms power much buying decisions, galore brands still don’t cognize whether they’re being recommended aliases overlooked. Prompt investigation addresses that uncertainty by focusing connected the moments wherever AI evaluates options and recommends a solution.

With Semrush, those determination moments go measurable signals you tin monitor, interpret, and enactment connected complete time.

Start by documenting 1 persona and generating 10 decision-stage prompts this week. Add them to Semrush's Prompt Tracking, past show wherever your marque appears, wherever it doesn't, and really AI frames your category.

From there, AI visibility becomes thing you tin activity with, not conjecture at.