Key takeaways
- Load the same 30 prompts into every tool, or you are comparing prompt sets instead of tools.
- Build the set from three prompt types: organic, brand-specific, and competitor comparison. Each pulls a different source profile.
- Run every trial in the same 7-day window so model and retrieval changes hit all tools equally.
- Score on repeatability, UI vs API data, citation detail, and what the tool helps you do next, not just the dashboard.
- Check trial caps first. Some trials will not fit 30 prompts, and a free tier or one-off audit is not a trial.
Why most tool comparisons are unfair
I have seen teams sign up for three AI visibility tools, type in whatever prompts came to mind in each, and then argue about which tool "shows more mentions." That comparison is meaningless. Different prompts, different days, different countries, different brand aliases. The tools were never tested against each other.
The fix is boring and it works: one prompt set, one time window, one scoring sheet. A comparison guide from Suparank says it plainly: load the same prompts into every tool you trial, otherwise you are comparing prompt sets, not tools. This guide turns that into a concrete 7-day plan.
Before you start: know what counts as a trial
DeepSmith makes a distinction worth keeping. A free trial is time-boxed, usually 7 or 14 days, and converts to a paid plan unless you cancel. A free tier is a standing allowance that never expires. A one-off free audit is neither. Mixing these up is how you end up "testing" a tool on three prompts and calling it a day.
That DeepSmith guide screens trials on public pricing, multiple AI engines, and sensible prompt counts, which is a decent filter for your shortlist too. Another published comparison notes a 3-day trial being shorter than the category norm, and a 7-day trial running 50 prompts a day on some tools. Trial terms vary a lot, so check these before you commit a week:
- Trial length (3, 7, or 14 days)
- Whether a credit card is required
- Prompt cap during the trial (will 30 fit?)
- Which engines are unlocked in the trial
- Whether data refreshes daily or weekly
The research I have does not verify these for every tool, so confirm them on each vendor's pricing page. Pricing also moves. Profound, for example, publishes no base price at all, only a free trial and a custom Enterprise quote, according to Rework's October 2026 comparison, while Scrunch and Goodie publish theirs.
Step 1: Shortlist two or three tools
Three is the practical ceiling for a 7-day trial. More than that and you will not have time to inspect the results properly. Pick tools from different categories so the trial teaches you something:
| Category | Example tools | What you learn |
|---|---|---|
| Monitoring-focused trackers | Otterly.AI, Peec AI, Goodie AI | Is basic mention tracking enough for you? |
| Enterprise platforms | Profound, Scrunch AI | Does a higher price buy better data? |
| Optimization platforms | Promptwatch | Does the tool help you act, not just observe? |
| SEO suites with AI add-ons | Semrush, Ahrefs Brand Radar | Is a bolt-on good enough? |
Otterly.AI
Profound
If you want a wider list to pick from, the GEO software directory at bestgeosoftware.com and the AI rank tracking directory at ai-rank-tools.com are good starting points.
Promptwatch is worth including as the optimization-focused option, since it tracks real UI answers and also covers crawler logs, visitor analytics, and content gaps, which most pure trackers skip. It has a 7-day free trial on monthly paid plans, but confirm current terms at promptwatch.com/pricing.
Step 2: Build the 30-prompt set
Thirty is enough to see patterns and small enough to fit most trial caps. Do not write them all in one sitting from your own head. You will bias them toward prompts you already win.
Split by prompt type
Promptwatch Data's unique domains per response report (Sep 28 to Oct 4, 2026) shows why this split matters. The source profile changes with the prompt type:
| Prompt type | Example | Share citing 10+ domains | Notes |
|---|---|---|---|
| Organic | "best project management software for small teams" | 52.5% | 71.8% cite at least 8 domains |
| Brand-specific | "is Asana good for small teams" | 21.3% | 5 domains is the most common count |
| Competitor comparison | "Asana vs Monday.com" | 49.1% | 66.5% cite at least 8 |
Across all ChatGPT responses with a citation, 47.6% cite 10 or more domains. With that many sources per answer, a single prompt tells you very little, which is exactly why you need a set. A sensible split:
- 14 organic prompts (no brand named)
- 8 brand-specific prompts (your brand and two or three variants)
- 8 competitor comparison prompts (you vs each main rival)
Where to source them
- Sales call questions and support tickets
- Google Search Console queries that sound like questions
- People Also Ask results for your core topics
- A mix of short and long prompts, since fanout query length shifts over time
Add tags to each prompt (type, topic, funnel stage) in a spreadsheet. You will use them when comparing results.
Freeze the set
Once written, do not edit. Lock the exact wording, country, and language. If a tool lets you add personas or locations, either use the same settings in all of them or leave them off. Every difference you introduce is a variable you cannot untangle later.
Step 3: Standardize everything else
Prompts are only half of it. These also need to match across tools:
- Brand name and aliases (misspellings, product names, abbreviations)
- The same competitor list, entered manually if a tool auto-discovers a different one
- Country and language
- Engines: pick the two or three that matter, such as ChatGPT, Google AI Overviews, and Perplexity
- The start day, so all trials share the same window
The same-window rule matters more than it sounds. Promptwatch Data shows engines change behavior mid-flight. ChatGPT started using the site: operator at scale on August 8, 2026, jumping from about 0.4% to about 17% of fanout queries (report), and average citations per response dropped around the GPT-5.3 rollout on March 4, 2026 (report). If you test tool A this week and tool B next month, you may be measuring the model change, not the tool.
Step 4: The 7-day schedule
| Day | Task |
|---|---|
| 1 | Sign up for all tools, enter the 30 prompts, brand aliases, and competitors. Screenshot every setting. |
| 2 | Confirm all prompts ran. Log setup time and any friction. |
| 3 | First data check. Compare raw answers for 5 prompts across tools. |
| 4 | Check repeatability: do the same prompts give similar results as day 2? |
| 5 | Test the secondary features: citations, reports, exports, integrations. |
| 6 | Run the same question against each tool's support or chat assistant, if it has one. Note response quality. |
| 7 | Score, decide, cancel what you will not keep. |
Set a calendar reminder to cancel before the trial converts. That is the single most common way trials turn into surprise invoices.
Step 5: What to actually measure
Repeatability
Engines differ in how noisy they are. Promptwatch Data's average sources per response report (an older page, so treat figures as dated) found Perplexity citing almost exactly 10 sources daily, which makes it a stable test bench, while Copilot swung from under 2 to nearly 17 within weeks. In a 7-day test, expect Perplexity results to be the most repeatable and Copilot the least. If a tool reports wildly different numbers for the same prompt on consecutive days on a stable engine, that is a signal about the tool or its sampling.
UI data vs API data
Ask each vendor whether it collects answers from the real user interface or from the API. User-facing answers and citations can differ from API outputs. Promptwatch, for example, states it collects UI data from ChatGPT, Gemini, Perplexity, AI Overviews and others. Whichever way a tool does it, you want to know, because it changes how much you can trust the numbers as a picture of what customers see.
Detail behind the score
A visibility percentage with no source list is a dead end. Check whether you can click through to the full response, the cited URLs, and the position of your brand in the answer. Promptwatch's own advice from that report is to track where you appear and how ChatGPT describes you, not just whether you are cited.
Setup and prompt accounting
How does each tool count prompts: per prompt, per prompt per engine, per country? A "50 prompt" plan can mean very different things. Count the real cost of tracking your 30 prompts on three engines in two countries.
What happens after you see the problem
This is where tools split. Trackers answer "was my brand mentioned?" Optimization platforms try to answer "what do I do about it?" During the trial, test whether the tool can show you why you are invisible (crawler logs, site audits), which pages get cited, and what to publish. If you only ever need monitoring, a cheaper tracker like Otterly.AI may be plenty. If you need to act on the data, look at tools that include content gap analysis and publishing, such as Promptwatch's Content Agent and Action Board.
Step 6: Score with a shared sheet
Use a simple weighted score. Adjust the weights to your situation, but decide them before you look at results, so you do not rationalize a favorite.
| Criterion | Weight | Tool A | Tool B | Tool C |
|---|---|---|---|---|
| Fit with your 30 prompts (all ran, right engines) | 15% | |||
| Repeatability across days | 20% | |||
| Data source (UI vs API) and transparency | 15% | |||
| Citation and source detail | 15% | |||
| Actionability (what it helps you fix) | 20% | |||
| Cost at your real prompt volume | 15% |
Score each 1 to 5, multiply by weight, and total. Then sanity-check: if the numerical winner feels wrong, find out why before you buy.
Common traps
- Letting each tool auto-generate prompts. You are back to comparing prompt sets.
- Judging by the headline visibility number. Different tools calculate it differently, so compare raw responses instead.
- Testing in different weeks. Model changes land on everyone at once, but only if you are testing at the same time.
- Ignoring who owns the tool. Scrunch was announced as owned by Sitecore on June 3, 2026, which may matter for roadmap and procurement.
- Forgetting that AI visibility tools cannot fix a site AI crawlers cannot read. Check crawlability with a free tool like the Agentic Readiness Check from Promptwatch before you blame the tracker.
After the trial
Pick the winner, keep the 30 prompts as your baseline, and re-run the same set monthly. That gives you a before-and-after view when you publish new content. If you would rather have a team handle the tool selection and the optimization work, 1001 SEO Media runs GEO programs for brands and uses Promptwatch as its platform.
For current research on how AI engines pick sources, the Promptwatch Data hub is free and updated regularly. Confirm all vendor pricing and trial terms on their own sites before signing up, since they change often.