Set up tracking

Choosing the right prompts

What makes a prompt worth tracking, how to mix buyer-journey stages and branded vs. unbranded questions, and how many prompts you actually need.

Open in ChatGPT Open in Claude

Your prompts are the sample every AmICited number is computed from. The visibility score, share of voice and every report downstream only describe the questions you chose to track. A good prompt set looks like the questions real buyers type into ChatGPT or Perplexity. A poor one produces flattering or meaningless numbers.

What makes a good prompt#

Written the way a buyer asks. “What’s the best CRM for a 10-person agency?” is a prompt. “CRM agency” is a keyword. AI engines answer full questions differently from fragments, so track the question.
One question per prompt. “Best CRM and how much does it cost” mixes two intents and gives you an answer you cannot act on. Split it.
Answerable with a recommendation. Prompts that invite a list of options (“best”, “top”, “alternatives to”, “which tool for”) are where brands get named. Pure definitions (“what is a CRM”) rarely name anyone.
Specific enough to be yours. Add the qualifier that makes you a real contender: an industry, a company size, a use case, a region. “Best CRM” is dominated by the biggest brands; “best CRM for construction subcontractors” is winnable.
Stable over time. Avoid dates or current events in the text unless that is the point. A prompt with “2025” in it will look stale next year.

Keep prompts short enough to carry keyword data too. AmICited looks up search volume, CPC and competition for each prompt text, but the lookup skips texts longer than about 10 words or 80 characters, so very long prompts show no volume.

Cover the buyer journey#

Spread prompts across the stages where AI answers influence a purchase. A useful starting mix:

Stage	Example prompt	Why track it
Problem	“How do agencies keep track of client follow-ups?”	Shows whether you are part of the conversation before buyers know the category.
Category	“Best CRM for small marketing agencies”	The core shortlist question. Usually the biggest share of your set.
Comparison	“HubSpot vs Pipedrive for agencies”	Tells you how engines position you against named rivals.
Alternatives	“Cheaper alternatives to Salesforce”	High intent, and often cites review sites and listicles you can get onto.
Brand	“Is Acme CRM good for agencies?”	Checks what engines say when buyers already know your name.

The Prompts page labels each prompt’s intent (Informational, Navigational, Commercial, Transactional, Local, Exploratory, Comparative research, Synthesis) and lets you filter by it, which makes it easy to check the balance of your set after you add it.

Branded vs. unbranded#

Unbranded prompts (no brand name in the text) are where visibility is won or lost: the engine decides whom to recommend. Branded prompts (“Acme vs HubSpot”, “is Acme legit”) almost always mention you, because you are in the question.

Keep branded prompts to a small minority, roughly 10 to 20% of the set, and tag them so you can filter them out. A set that is half branded will show an inflated visibility score that does not reflect how buyers discover you.

Tag by stage from day one

Create tags such as category, comparison and branded as you add prompts. Every dashboard figure can then be filtered to one stage, which is how you spot that you win comparisons but miss category questions.

How many prompts you need#

More prompts give steadier numbers, but each prompt costs credits on every run. Rough guidance:

20 to 30 prompts is enough to see where you stand in one product category. With 25 prompts, one prompt moves the visibility score by 4 points, so expect some noise.
50 to 100 prompts gives stable trends for a single brand across several use cases or segments.
Hundreds of prompts make sense for agencies, multi-product brands or several countries, usually split with tags.

Budget before you scale. A prompt tracked daily on the five standard engines uses about 2.25 credits a month; the same prompt weekly uses about 0.32. Run frequency has the full math, and your plan’s prompts-per-domain cap is on plans and limits.

Where to find prompt ideas#

Suggested from your site. The setup wizard and the Generate from URL method read a page and propose questions buyers are likely to ask.
Keywords you already rank for. Both flows also list the keywords a URL ranks for on Google, with volume, CPC and position, so you can turn proven demand into prompts.
Your competitors’ keywords. Adding a competitor by URL can pull the keywords they rank for. See competitors.
Your own search data. Once Search Console is connected, question-style queries from your organic traffic are strong candidates.
Sales and support. The questions prospects ask on calls are the questions they ask AI.

Review the set every quarter#

Remove prompts that never name any brand (they measure nothing), merge near-duplicates, and add questions for new products or markets. Change the set deliberately: adding 20 prompts at once shifts your averages even if nothing else changed. When you compare periods, compare them with the same tag filter.

Adding prompts Paste, upload a CSV, or generate prompts from a URL. Tags Group prompts by stage, product or market. Run frequency How often each prompt runs, and what it costs. Not-ranked prompts The prompts where you are missing, and what to do next.