Core feature
Track the questions your buyers actually ask
Every buyer question is a prompt, and every prompt gets a different answer. Siftly tracks your prompt set across ChatGPT, Perplexity, and Google AI Overviews, samples each one repeatedly, and shows whether your brand was named, which competitors were named beside you, and which sources the answer was built from.
the prompt set
Your prompt set is your keyword list, and it moves
Start from the questions your buyers ask. Onboarding generates a starting set from your domain, your topics, and your personas. You edit it, add your own, and archive the ones that stop mattering. Each prompt carries the topic and personas it belongs to, so results roll up the way your team thinks.
One answer proves nothing. Engines are non-deterministic: the same question can name you in one run and skip you in the next. Siftly runs each prompt repeatedly and reports the rate, not a single lucky or unlucky snapshot.
every execution
Read the answer, not just the score
Open any run and see what was said. Each execution keeps its platform, date, and location, the brands the answer named, the search queries the engine fanned out to, and the citations it built the answer from. Your own brand is marked in the mention list, so presence and absence are both obvious.
Find the prompts you are not tracking yet. Siftly proposes new prompts from two places: answers where a competitor is cited and you are not, and searches you already appear for in Search Console. Each suggestion carries its intent and its source, and one click adds it to your tracked set.
AI Prompt Tracking
What AI Prompt Tracking Measures
How it works
How Sampling Works
Run each prompt repeatedly
A single answer is one sample of a probabilistic system. Siftly runs your prompts on a daily schedule so the result is a rate across many answers rather than one draw.
Run them per engine
ChatGPT, Perplexity, and Google AI Overviews answer the same question differently. Each engine is sampled separately, so you can see where you are strong and where you are absent.
Run them per location
Answers change by country. Prompts carry the locations you track, and every execution records the one it ran in, so a global average never hides a market where you are invisible.
Report the rate
Mention rate is the share of runs that named your brand. A brand in four answers out of ten has a 40% mention rate, which a single-shot checker would report as either 0% or 100%.
AI Prompt Tracking
What Each Execution Records
| What is recorded | Why it matters |
|---|---|
| Platform and date | Engines drift. Dated, per-engine records are what let you attribute a change to a model update rather than to your own work. |
| Location | The same prompt answers differently by country, so an execution without a location is not comparable to another. |
| Mentioned | Whether your brand appeared in this specific answer, which is what the mention rate is built from. |
| Brand mentions | Every brand the answer named, with yours marked, so you can see who you are being shortlisted against. |
| Search queries | The queries the engine fanned out to before answering, which tell you what it thought the question was about. |
| Citations | The pages the answer was built from, which is where content and outreach work starts. |
AI Prompt Tracking
Where New Prompts Come From
Competitor citations
Questions where a rival's page is the source and yours is not
Search Console
Searches you already rank for that engines are now answering
Intent
Commercial, evaluation, informational, or navigational
One click
Accept a suggestion and it joins your tracked set
Getting started
Time to value, not time to configure
Day 1
Generate the set
Siftly reads your domain and builds a starting prompt set across your topics and personas. Review it, edit anything, and add the questions your sales team hears.
Day 2
First full sweep
Every prompt runs across your entitled engines and locations. You get mention rate, the competitors named beside you, and the sources each answer used.
Week 2
Work the gaps
Filter to the prompts where you are missing, read the answers that beat you, and send them to content or outreach. Review the suggestions queue as it fills.
Questions