The AI search prompt library for GEO research.
Want to know what ChatGPT, Perplexity, Gemini, Copilot and Google's AI Mode say about your brand? Start by asking the right questions. These 65 copy-ready prompts are the ones our GEO team uses to audit brand perception, map competitors, find the sources AI cites and build a prompt set you can track every week.
with variables
from audit to tracking
them on
You can't optimise for a question you never asked.
In classic SEO, you research keywords. In generative engine optimisation, you research prompts — the full, conversational questions buyers type into AI assistants.
Prompts are longer, more specific and more personal than keywords: "best accounting software for a 10-person agency that invoices in USD" rather than "accounting software". Each one produces a different answer and a different list of brands. The only way to understand your AI visibility is to run a representative set of these questions, repeatedly, across engines — and score what comes back.
- Variables in orange — replace [brand], [category], [city] etc. with your own details.
- Copy button on every prompt, so you can paste straight into any AI tool.
- Organised by purpose — from a quick perception check to a full tracking set.
65 prompts, organised by what you need to learn.
Choose a tab. Start with brand perception and competitor comparison for a quick health check; use the later tabs to build a full audit and a tracking set.
Purpose: learn how AI engines describe your brand unprompted — strengths, weaknesses, positioning and tone. Run these first; they reveal what the model "believes" before any comparison.
Purpose: see how you're positioned head-to-head and which competitors AI considers your peers. Note who gets recommended "by default" when the user expresses no preference.
Purpose: the money prompts. These "best X for Y" questions create short lists. Run many variations — budget, size, industry, region — because visibility is often strong in one variation and absent in the next.
Purpose: local intent behaves differently — answers lean on maps data, reviews and local directories. Run these from the target location (or with the city named explicitly) and compare. Pairs with Google's AI features — see AI Overviews optimisation.
Purpose: product discovery inside AI assistants and shopping experiences. Note which products are named, whether price and availability are right, and which retailers or reviews are cited.
Purpose: B2B decisions involve several people asking different questions. Write prompts from each role's perspective — the CFO's question is not the end user's question.
Purpose: find the pages AI engines lean on for your category — the review sites, listicles, forums and publications that shape answers. These become your outreach and PR target list. Best run on engines that show sources, such as Perplexity and ChatGPT search. Read more on AI citation optimisation.
Purpose: mine the questions AI assistants expect around your topic — then make sure your site answers them better than anyone. These prompts feed your content calendar and FAQ blocks.
Purpose: catch wrong facts before buyers do — outdated pricing, wrong locations, discontinued products, invented features. Compare every answer against your source of truth and log errors. Deeper context: AI brand mentions.
Purpose: use AI to help draft your tracking prompt set — then edit it with your sales team. A good set mixes unbranded and branded prompts across funnel stages, personas and regions.
All prompts are generic templates for research. Replace the orange variables with your own details. Results will vary by engine, account, location and date — that's why the run method below matters.
Don't have time to run them? We'll run 200 prompts for you.
We'll build a prompt set from your buyers' real questions, run it across ChatGPT, Perplexity, Gemini, Copilot and Google AI Mode, and show you exactly where you win and lose — with a scored spreadsheet you keep.
Six rules for prompt research you can actually trust.
One run in your own logged-in account tells you almost nothing. AI answers vary from run to run, user to user and place to place. Here's the method we use.
Neutralise personalisation
Use logged-out sessions or fresh accounts with memory and custom instructions off, in a private browser window. Your own history — including all the times you've searched your brand — biases answers.
Run each prompt several times
Run every prompt at least 3–5 times per engine and record each result. Report a mention rate ("named in 3 of 5 runs"), never a single yes/no.
Cover multiple engines
ChatGPT, Perplexity, Gemini, Copilot and Google AI Mode/Overviews use different retrieval and sources. Being visible in one says little about the others.
Control location & language
For local and regional prompts, run from the target location (or a VPN set there) and in the buyer's language. Note the location on every record.
Record everything, the same way
Date, engine, model/mode, prompt, full answer text, brands named, position, cited URLs, accuracy and sentiment. Screenshots are evidence, not data.
Keep the set stable
Change your tracked prompts rarely and deliberately. If the questions change every month, you can't tell whether visibility moved or the test did.
Why multiple runs aren't optional
In research published in January 2026, SparkToro and Gumshoe had 600 volunteers run 12 prompts a combined 2,961 times across ChatGPT, Claude and Google's AI. They found less than a 1-in-100 chance that ChatGPT or Google's AI would return the same list of brands in any two responses — and recommended measuring visibility percentage across many repeated runs instead of "rank". Source: SparkToro, 2026
Practical takeaway: a single screenshot of "we're #1 in ChatGPT" proves very little. Tens of runs per prompt cluster, recorded consistently, prove a lot.
Turn answers into data with one simple sheet.
Copy these columns into a spreadsheet. One row per run. After a few dozen rows, patterns jump out: which engines ignore you, which competitors dominate, and which facts AI keeps getting wrong.
| Column | What to record | Example value | Scoring |
|---|---|---|---|
| Date / engine / mode | When and where you ran it | 2026-10-05 · Perplexity · default | — |
| Prompt ID & cluster | Stable ID from your prompt set | DISC-15 · "best X for SMB" | — |
| Location / language | Where the run was made from | Mumbai · English | — |
| Mentioned? | Is your brand named at all | Yes | 1 = yes, 0 = no |
| Position | Order among brands named | 2 of 5 | 1st = 3, top 3 = 2, other = 1 |
| Recommended? | Named as a recommendation, not just listed | "Strong option for…" | 1 / 0 |
| Cited URLs | Every source shown, flag your own | yourbrand.com/pricing; review-site | Own citation = 1 |
| Accuracy | Facts checked against source of truth | Old price quoted | Error = −1, log it |
| Sentiment | Tone of the description | Positive | +1 / 0 / −1 |
| Competitors named | All other brands, in order | Competitor A, Competitor B | For share of voice |
Mention rate
Runs where you're named ÷ total runs, per engine and per cluster.
Your headline visibility number.
Share of voice
Your mentions ÷ all brand mentions (you + competitors) in the same runs.
How much of the conversation you own.
Citation share
Runs citing your URLs ÷ runs with citations.
Which pages earn trust from AI.
How a prompt set becomes a growth dashboard.
Prompts aren't a one-off exercise. Once scored, they become the backbone of ongoing AI visibility tracking.
Draft & validate
Generate candidates with the tracking-set prompts, then validate with sales calls, site search, Search Console queries and customer interviews.
Cluster & weight
Group into clusters (use case, persona, region) and weight by commercial intent so a "pricing" prompt counts more than a definition.
Baseline
Run the full set across engines with repeated sampling. This is your "before" picture.
Act on gaps
Lost prompts become content briefs; cited third-party sources become PR targets; accuracy errors become entity fixes.
Re-run on a schedule
Weekly or monthly runs show trends by cluster and engine, tied to AI referral traffic in analytics.
1. Competitor A — popular with small agencies, strong multi-currency support.[1]
2. yourbrand — simpler interface, good for teams without a finance lead.[2]
3. Competitor B — best if you need deep project costing.[3]
Illustrative example. Scored: mentioned = 1, position = 2, recommended = 1, own citation = 1, sentiment = +1.
What separates useful prompt research from noise.
The golden rule
Measure with neutral prompts, act on what you learn, then measure again with the same prompts. Research and optimisation are separate steps — mixing them is how teams fool themselves.
How we test changesDo
- Write prompts the way buyers talk — long, specific, with context.
- Include unbranded prompts; that's where new customers discover you.
- Track 3–5 named competitors so share of voice means something.
- Keep a "source of truth" sheet of facts to check accuracy against.
- Segment results by engine, persona and funnel stage.
- Re-run the same set over time and annotate site changes.
Don't
- Judge visibility from one run in your personal account.
- Only test prompts that contain your brand name.
- Use leading prompts ("why is [brand] the best?") for measurement.
- Change the prompt set every month.
- Try to "train" the model by repeatedly praising your brand in chats.
- Publish hidden text or prompt-injection tricks aimed at AI crawlers.
Prompt research is where GEO starts. We take it to revenue.
A prompt library tells you where you stand. Our team turns that into the content, entity work and digital PR that changes the answers.
- Prompt sets built with your sales team — 50–200 real buyer questions, weighted by intent.
- Repeated, multi-engine sampling across ChatGPT, Perplexity, Gemini, Copilot and Google AI features.
- Source maps that turn citations into a prioritised PR and content plan.
- Full-stack execution — SEO, content, PR and dev in-house, around 300 specialists, since 2008.
See what happens next in our GEO engagement playbooks, or compare the AI SEO tools that automate tracking.
Included in our "200 prompts" audit
Get your AI visibility baseline this week.
Tell us your category and top competitors. We'll run your prompt set and walk you through the results on a 30-minute call.
AI search prompt questions, answered.
What is an AI search prompt library?
It's a structured collection of questions you run on AI assistants such as ChatGPT, Perplexity, Gemini and Copilot to understand how they describe, compare and recommend your brand. In GEO, prompts play the role keywords play in traditional SEO research.
How many prompts do I need to track AI visibility?
For a single product line, 50–200 prompts usually gives a representative picture across funnel stages, personas and regions. Fewer than 30 tends to be too noisy; far more than a few hundred is rarely needed unless you have many products or markets.
Why do I get different answers each time I run the same prompt?
AI answers are generated probabilistically and may draw on different live sources each time. Personalisation, location and model updates add more variation. That's why you should run each prompt several times and report a mention rate rather than a single result.
Should I run prompts logged in or logged out?
For baseline measurement, use logged-out or clean accounts with memory and custom instructions turned off, so your own history doesn't bias results. It can be useful to run a few logged-in checks separately, as many real buyers use personalised accounts.
Which AI engines should I test?
At minimum ChatGPT, Perplexity, Google's AI features (AI Overviews and AI Mode), Gemini and Microsoft Copilot. Prioritise based on your audience — B2B buyers and consumers often lean on different tools.
Can I use these prompts to change what AI says about my brand?
No. Asking or praising your brand in chats doesn't retrain public models. These prompts are for measurement and research. Changing answers requires work on your website, entity data, content and third-party mentions — which is what our GEO service does.
How often should I re-run my prompt set?
Weekly for active campaigns or competitive categories, monthly for most brands. Keep the core set stable so changes over time reflect real shifts in visibility rather than changes to the test.
Can tools automate prompt tracking?
Yes. A growing number of AI visibility tools run prompt sets on a schedule and report mention rate, share of voice and citations. They save time, but the prompt set itself still needs human judgement. See our comparison of AI SEO tools for options.
Keep exploring generative engine optimization
Every AI platform rewards slightly different signals. Dive into the playbook for each one — or let our GEO team run all of them for you.
Find out what ChatGPT, Gemini & Google AI say about you today.
Tell us your website and top competitors. Within 3 business days a GEO strategist sends you a personal AI visibility report — no generic PDF, no sales script.
- Share-of-voice snapshot across ChatGPT, Gemini, Perplexity, Copilot and Google AI Overviews
- Prompt-level gaps where competitors are named and you aren't
- Citation sources each AI engine leans on in your category
- A prioritised 90-day GEO plan you can run in-house or with us