If you lead content, you've probably been asked how your pages show up in AI answers, and you may not have a clean way to brief, check and track that work. This guide gives you three GEO templates you can copy: a content brief, a page audit and a starter prompt set. Each one comes with notes on how to fill it in, so by the end you'll have a full set of records for one page.
Before you start, it helps to know what these are and what they aren't. There is no official or standardized template for this. The fields below are practical editorial recommendations, drawn from published examples and from what's worth writing down, and none of them is a score that predicts whether a page will be cited. They give you a consistent record of your buyer questions, your evidence, the condition of the page and what AI engines actually said.
You'll need one page in mind, either an existing URL or one you're about to brief. The work goes in seven steps, and each says what to fill in, how to tell it's done and where people go wrong.
Step 1: Define the page's job
Start with a few plain facts about the page: the page type, the audience, the buyer stage, the decision or task the reader is trying to get through, and the primary topic. Name the reader and the problem the page is meant to solve. For a marketing lead that might be understanding a category, comparing approaches or choosing a tool.
You're done when someone else on the team could say in one sentence who the page serves and what useful outcome it should give them.
The common mistake is making "rank in AI" the only goal. It doesn't tell the writer what the reader needs or what the page has to answer. If you want the wider background, our plain-English definition of generative engine optimization is a good place to read first.
Step 2: Map the buyer questions
Next, write down a small but varied set of questions in the words buyers really use. Mark each one with its intent and buyer stage, and keep the broad learning questions apart from the comparison and decision questions. Copy the exact wording instead of squeezing everything into keyword fragments, because the exact wording is what you'll reuse in the prompt set later.
You're done when the questions cover more than one intent and you could build a real page outline from them. Support tickets and sales calls are a good source, and there's a walkthrough on how to mine support tickets and sales calls for the questions buyers ask. If you're starting from nothing, this guide to finding the prompts to track shows where to look.
The common mistake here is writing only brand-led questions such as "Why is [Brand] the best?" That kind of question is leading, and it doesn't tell you whether your brand shows up in an unbranded discovery answer.
Step 3: Fill in the GEO content brief
Here is the GEO content brief template. Copy the block below for one page. This GEO content brief template covers the main question, the answer the page must give, supporting questions, evidence, entities, structure, exclusions and a success measure.
GEO CONTENT BRIEF
A. PAGE BASICS
Working title:
Proposed URL (if known):
Page type:
Brief owner / writer / reviewer:
Brief date:
B. AUDIENCE AND JOB
Primary audience/persona:
Buyer stage or user situation:
What the reader is trying to do:
Page goal:
What the reader should understand or be able to do after reading:
Primary topic:
C. QUESTIONS TO ANSWER
Main question, in the reader's words:
Short, direct answer the page must provide:
Supporting question 1:
Supporting question 2:
Supporting question 3:
Important comparison or decision question, if relevant:
Questions deliberately out of scope:
D. PROMPT AND INTENT MAP
Exact buyer prompt:
Intent: learn / compare / evaluate / decide / other:
Audience or use case:
Why this prompt belongs on this page:
Related prompt or wording variant:
[Repeat for each selected prompt. Keep wording exact if it will later be monitored.]
E. FACTS, EVIDENCE, AND CLAIMS
Fact or claim to include:
Source / evidence owner:
Date checked or "as of" date, if time-sensitive:
What the evidence establishes:
Caveat, qualification, or context the writer must retain:
Claims to avoid or verify before publishing:
Original examples, data, expertise, or first-party material the page can add:
F. ENTITIES AND TERMINOLOGY
Company, product, person, place, or category names to define:
Preferred spelling and description:
Terms readers may confuse:
Definitions or distinctions to make explicit:
G. CONTENT PLAN
Proposed H1:
Answer or summary to place near the top:
H2 / section 1:
H2 / section 2:
H2 / section 3:
Useful example, table, diagram, or media:
FAQ questions, only if they address genuine reader needs:
Relevant existing pages or internal-link candidates for the linking editor:
H. QUALITY AND TECHNICAL NOTES
Important information available as visible text: yes / no / check
Page is intended to be indexable: yes / no / check
Canonical / duplicate-page concern, if known:
Relevant structured data, if applicable:
Structured data must match visible page content: yes / not applicable / check
Platform-specific crawl or snippet controls to check, if relevant:
I. SUCCESS AND REVIEW
Primary reader outcome:
Baseline and measurement source:
Platform, metric, and date range to use:
Success signal, defined without a promised citation outcome:
Review date or trigger:
A few notes on filling in the brief. Write the main question the way the reader would say it out loud, and put it ahead of any SEO shorthand. The "direct answer" field means a useful answer near the top of the page, not a mandatory word count, and Google doesn't specify an answer length for its AI features.
Keep the evidence field claim by claim. A general list of sources at the bottom doesn't make an unsupported statistic checkable. Ask for things the team can really back up, such as firsthand experience, original data or an accurate product detail, and don't label an opinion as a verified fact.
Add schema only when the page visibly contains the information the markup describes. Google says it has no special schema for its generative AI search features, so don't write an "AI schema" requirement into the brief. For success, use a reader outcome plus something you can observe, like an indexed page or a recorded citation on a named prompt and engine.
You're done when the writer knows what to say, what supports it, what not to claim and how the page serves the reader. The common mistake is treating "AI-friendly" structure as a replacement for original evidence or a genuinely useful answer.
Step 4: Set evidence and accuracy rules
Before anyone drafts, decide how facts get handled. For each factual claim, note where it comes from, who owns or reviews it, whether it has a date that can go stale, and what the page is not allowed to say. Ask for first-party records or authoritative sources where they exist, and decide up front how estimates, projections and open questions will be treated.
You're done when every material claim has a source or a plain "verify before publishing" note next to it.
The common mistake is letting a drafting prompt invent customer results, statistics, quotes or product capabilities so the article sounds authoritative.
Common mistake: Treating one answer from one engine as proof of a stable citation position. AI answers and sources can vary across prompts, runs and time. Record the exact prompt, platform, date and page cited, and label one-off observations as snapshots.
That's worth taking seriously. A 2026 repeated-measurement study ran the same prompts across four engines and found that source sets overlapped only 34 to 42 percent across consecutive days, and brand sets 45 to 59 percent. The authors measured visibility repeatedly and suggested averaging over several runs and a few weeks. Those numbers come from one study, using Swiss servers, so they aren't thresholds for your business. The takeaway is to write down the exact prompt, engine, date and run.
Step 5: Record page findings with the GEO audit template
The second of the GEO templates is the page record. This GEO audit template works for an existing or proposed URL. It's a list of evidence to write down, and it isn't a set of instructions for running a technical crawl or check. If you want the step-by-step version of checking a page, that's covered in how to run a GEO check on a page.
GEO PAGE AUDIT
A. RECORD DETAILS
Page URL:
Page title:
Audit owner:
Observation date:
Page type / topic:
Target audience and stage:
Primary intended prompt or question:
Locale / language:
B. PAGE CONDITION
Canonical URL or canonical status, if known:
Indexing status / evidence / unknown:
Robots or other crawl controls relevant to the target platform:
Important content available in visible text: yes / no / unknown
Internal discovery/link note:
Redirect or duplicate URL note:
Last updated / author, where relevant:
C. CONTENT USEFULNESS
Does the page answer its main question clearly? yes / partly / no / unknown
Is the answer easy to find near the beginning? yes / partly / no / unknown
Does the page satisfy the intended reader's need without missing context? yes / partly / no / unknown
Original evidence, expertise, examples, or useful detail present:
Facts that need verification or an "as of" date:
Entities and product names clear and consistent? yes / partly / no / unknown
Unsubstantiated, vague, or overly broad claims:
Content or sections that are off-topic, duplicative, or unnecessary:
D. STRUCTURE AND SUPPORT
Clear title and headings: yes / partly / no / unknown
Definitions, comparisons, or steps explicit where needed: yes / partly / no / unknown
Images/video supported by text, captions, alt text, or transcript where relevant:
Structured data present / relevant / absent / unknown:
If present, does it match visible content? yes / no / unknown / not applicable
FAQ present only when genuinely useful? yes / no / not applicable
E. AI ANSWER OBSERVATIONS
Prompt ID and exact wording:
Engine / experience:
Language / location, if known:
Observation date and time:
Run number or aggregation window:
Brand mentioned? yes / no
Brand cited as a source? yes / no
Cited page URL(s), if any:
Competitor or other cited source, if relevant:
Answer excerpt or evidence note:
Record source: direct observation / platform report / other
[Repeat this block per prompt, engine, and observation. Do not merge unlike platforms or prompt wording.]
F. FINDINGS AND NEXT ACTION
Status per finding: Pass / Needs work / Unknown / Not applicable
Finding and evidence:
Why it matters to the reader or measurement:
Recommended next action:
Owner:
Priority: Now / Later / No action, with reason
Recheck date or condition:
G. SUMMARY
Most useful thing the page already does:
Most important evidence or accuracy gap:
Most important reader-value gap:
Observed AI visibility signal, with platform and period:
What this audit cannot establish:
Some notes on using it. The statuses are Pass, Needs work, Unknown and Not applicable, and they are descriptions only. The original GEO research paper is a good reminder of why. It reported its gains on its own 10,000-query benchmark, so those figures describe that experiment and not a lift you should expect for a business page. They aren't weighted points, and they don't imply a search-engine score. Keep what you notice about the page separate from what you notice about AI answers, because a well-written page and a cited page are two different observations.
For every answer observation, keep the exact prompt, engine, date and cited URL. If a platform only gives you aggregated reporting, write down that limit instead of implying you saw every answer. When you don't know whether a page is crawlable or indexed, write "unknown" and don't guess from how the page looks.
It also helps to know what platform reports can tell you. Google Search Console's generative AI performance report counts impressions for links shown in its generative AI features, and Bing Webmaster Tools has an AI Performance report with sampled, aggregated citation activity whose grounding-query rows are short grouped phrases, not full prompts. Both are good for trends. Neither is a complete log of every answer, and a missing row doesn't prove a page was never cited. Keep the platform and metric next to each number, as mentions, citations and impressions mean different things.
You're done when each finding has a status, an evidence note and either a next action or a reason none is needed. The common mistake is giving the page a "GEO score" without saying what it means or what supports it. Existing audit examples, like the one in this GEO content audit template, include things like an inventory, goals and custom fields, and some suggest an answer under a set number of words. Those are suggestions and not validated thresholds, so don't copy a word count into your audit as though it were a rule.
If you'd like a fuller factor list to cross-check against, the AEO audit checklist is a good companion.
Step 6: Build and save the starter prompt set
The third template is a starter prompt set. Use the questions as patterns, replacing the bracketed parts with your real category, audience, use case and alternatives. The twelve prompts are an editorial starter set. They aren't a validated ideal count, and they aren't representative of every business, so treat them as a place to begin.
Keep the unbranded discovery prompts separate from the brand-specific ones. They answer different questions: the first group shows whether your brand enters a category answer, and the second shows how an engine describes you once it's named.
PROMPT SET RECORD
Brand / site:
Category:
Audience / persona:
Market and language:
Primary problem or job:
Relevant alternatives / competitors:
Prompt owner:
Prompt-set version and date:
PROMPT 01
Exact prompt: What is [category], and what problems does it solve for [audience]?
Intent: Learn / category discovery
Brand-led or unbranded: Unbranded
PROMPT 02
Exact prompt: What are the main ways [audience] can solve [problem]?
Intent: Learn / solution discovery
Brand-led or unbranded: Unbranded
PROMPT 03
Exact prompt: What should [audience] look for when choosing a [category] solution?
Intent: Learn / evaluation criteria
Brand-led or unbranded: Unbranded
PROMPT 04
Exact prompt: What are the advantages and disadvantages of [approach A] versus [approach B] for [use case]?
Intent: Compare approaches
Brand-led or unbranded: Unbranded
PROMPT 05
Exact prompt: Which [category] options are suitable for [specific audience or use case], and why?
Intent: Compare / shortlist
Brand-led or unbranded: Unbranded
PROMPT 06
Exact prompt: How does [decision factor, such as setup, integration, or support] affect the choice of a [category] solution?
Intent: Evaluate
Brand-led or unbranded: Unbranded
PROMPT 07
Exact prompt: What questions should a [role or audience] ask before buying a [category] product?
Intent: Evaluate / purchase readiness
Brand-led or unbranded: Unbranded
PROMPT 08
Exact prompt: Compare [Brand] with [Competitor] for [use case]. What are the important differences?
Intent: Brand comparison
Brand-led or unbranded: Brand-specific
PROMPT 09
Exact prompt: What are the strengths and limitations of [Brand] for [audience or use case]?
Intent: Brand evaluation
Brand-led or unbranded: Brand-specific
PROMPT 10
Exact prompt: Does [Brand] support [specific capability or requirement], and what evidence explains the answer?
Intent: Feature / evidence check
Brand-led or unbranded: Brand-specific
PROMPT 11
Exact prompt: Which [category] providers offer [specific requirement] in [market], and how do they compare?
Intent: Local or constraint-based comparison
Brand-led or unbranded: Unbranded
PROMPT 12
Exact prompt: What is the best next step for a [audience] that wants to [specific outcome] but needs to address [constraint]?
Intent: Decision / next step
Brand-led or unbranded: Unbranded
FOR EACH OBSERVATION
Prompt ID and exact version:
Engine / AI experience:
Market / language:
Date and time:
Repeated run or reporting period, if applicable:
Brand mention: yes / no / unclear
Brand citation: yes / no / unclear
Exact cited URL(s), if visible:
Other notable source or competitor:
Short answer evidence note:
Data source and known reporting limits:
Here's how to adapt them. Replace vague fields with your actual problem, audience, constraint, category and market. Make each prompt test one need, and split any prompt that mixes unrelated questions. Use competitor comparisons only where your readers really compare those options.
If you're new to the idea, what prompt tracking is covers how a prompt list turns into visibility numbers. The best prompt sets come from real buyer language, so pull from customer questions, sales conversations, search data and support requests. It's also worth keeping versions. If you change the wording in a material way, treat it as a new prompt and don't compare it with the old one as though nothing changed. For more on this, the guide to setting up a repeatable prompt set goes into how to run one over time.
This is also the step where a tool can take over the tracking. In DeepSmith's AI Visibility area you can enter the buyer questions you defined, or use Discover Prompts to suggest candidates from your product, persona and buyer-stage context. DeepSmith then checks those prompts on a schedule and keeps the answers, with mentions and citations reported separately. The Pages view shows which of your pages get cited and which prompts drive those citations. The template stays useful as the portable record of what each prompt is for and why you chose it.
You're done when the prompts cover several buyer intents and are specific enough to be repeated the same way next time. The common mistake is changing the wording between observations and then treating the results as a clean comparison.
Step 7: Save a baseline and choose one next action
Last, write down what you saw: the date range, the platform, the prompt version, whether you saw a mention and a citation, the cited URL, and whether the evidence is a single observation or a repeated or aggregated view. Then pick one next action: improve the brief, revise the page, clarify the evidence, or gather more observations before changing anything.
You're done when a future reviewer could tell what was observed, when, on which engine and prompt, and what the team decided to do.
The common mistake is calling a single answer a stable ranking, or taking it as proof the page will be cited again. A baseline from one afternoon is a starting snapshot and not a result, and the piece on why AI citations keep changing explains what drives that. If you later want to track how the numbers move, the measure AI visibility manually guide shows a low-cost way to do it.
One more point about eligibility, since the audit asks about it. Google says its established SEO practices still apply to AI Overviews and AI Mode, and that a page needs to be indexed and eligible to show in Search with a snippet. Its page on AI features and your website is worth a read and it says nothing is guaranteed. Bing's guidance likewise says GEO doesn't guarantee grounding or citations. Platforms differ, so check each one's current guidance, including crawler controls such as OpenAI's separate search and training crawlers, and keep platform-specific controls out of the fields you treat as universal.
What to do next
Pick one page and fill in all three records with your own evidence. Once you've got a baseline, run the same prompts again in a couple of weeks and compare them to what you wrote down. A good place to take your tool choices from here is our roundup of the best GEO tools, and for the broader picture of tracking, see the best AI visibility tools.
Nothing in these templates guarantees a citation. What they give you is a record you can repeat and hand to someone else. If you'd like a platform to track the prompts and show you which of your pages get cited, you can start a DeepSmith free trial and see your own prompts, answers and cited pages in one place.
Related guides
- What to Look For in GEO Tool Integrations: CMS, Search Console and Analytics: Check which CMS, Search Console and analytics connections matter, and which integrations add little for your stack.
- An AEO Reporting Dashboard for Marketing Leads: Metrics and Layout: Design a dashboard for marketing leads with the right metrics and layout.
- Where Brand Monitoring Data Lives: Exports, Dashboards and Reporting Hooks: Find where monitoring data can leave the tool: exports, dashboards and reporting hooks.
- AI Overviews vs AI Mode: What Each Means for Tracking: Compare AI Overviews and AI Mode on how each is triggered, personalised and cited, and decide what that means for sampling queries, repeat runs and reading results in.
- What Counts as a Meaningful Change? Alert Thresholds for AI Visibility: Choose thresholds that separate real shifts from run-to-run noise, with sample numbers to adapt.
