An occasional AI visibility check can cost $0; groas tracking costs $19 a month flat, plus any assistant subscription and your review time. The expensive part is asking the same questions again: the manual program below takes an illustrative 6.4 to 16 hours a month just to collect the answers.

That is the bill I would price before buying another dashboard. Finding out whether ChatGPT can name a brand today is one job. Building a record that shows whether buyers keep seeing it is another. The cheapest workable setup depends on which answer you need.

The cost breakdown: cash, setup, and recurring work

For a concrete comparison, take 12 buyer prompts, four engines—ChatGPT, Claude, Gemini, and Perplexity—and four rounds across a month. That is 12 × 4 × 4 = 192 checks. This is an illustrative monitoring program, not a claim about what every business needs or what a particular subscription covers.

Line itemWhen you payCost or timeSource of the figure
Manual spot checksPer check$0 in tool spend where free access is available; under five minutes for a basic checkManual-check overview
Writing 12 buyer promptsOne-off setupRoughly one hourPlanning assumption; replace with your own estimate
Collecting 192 answers manuallyRecurring, monthly6.4 hours at two minutes per check; 16 hours at five minutesArithmetic from the illustrative program
groas UnlimitedRecurring, monthly$19 flat for the whole teamgroas pricing
Assistant accessRecurring, if needed$0 on a suitable free tier; ChatGPT Plus is $20 a monthChatGPT pricing
Interpretation, comparison, and filingEvery roundYour time; separate from collectionDepends on the questions and depth of review
Writing, approval, and publishingWhen you act on a findingSeparate hours, separate ownerDepends on the page change

The range comes from the collection process, not a mysterious industry benchmark. Two minutes per check assumes a quick pass through opening a clean chat, pasting a prompt, waiting, and logging the result. Five minutes allows more room for saving the full answer, citations, screenshots, and competitor details. Neither estimate includes interpreting the month’s results.

Keep the hourly rate editable too. At an illustrative $40 an hour, 6.4 hours costs $256 in labor. At $75 an hour, it costs $480. Those are planning examples, not market rates. The check count stays the same; the bill changes with whoever does the work.

I would separate the invoice into three piles: setup once, collection every round, and decisions afterward. A subscription price only answers part of the cost question.

What $0 buys: an answer worth writing down

The free route does exactly what it says. I can open a fresh chat without history, ask a buyer question, and inspect the answer. Perplexity’s numbered source list, Google AI Overviews, and Gemini with grounding offer other places to check. A basic pass can cost $0 and take under five minutes.

A one-off tool such as AmIPicked follows the same split: a free, no-signup check covering ChatGPT and Gemini, with daily tracking held back as the paid step. That makes the distinction fairly plain. Free buys an observation, not a maintained record.

Before keeping an answer, I separate three things:

  • Mentioned: the brand is named.
  • Cited: the brand’s page is linked as a source.
  • Recommended: the brand is suggested as something to buy.

Those are not interchangeable. A brand can appear in an answer without supplying its evidence or winning the recommendation. Share of voice measures mentions against competitors on the same questions; it does not settle the other two questions.

For a useful manual baseline, I would run 10 to 20 real buyer prompts and record those three outcomes. Alongside them, I keep five fields:

  1. Exact prompt text.
  2. Assistant used.
  3. Date.
  4. Full answer.
  5. Every cited URL.

Without that record, a screenshot becomes trivia by next Tuesday. I cannot tell whether the wording, the engine, or the week changed the result.

Assistant access need not add a bill at this scale. ChatGPT offers Free at $0 and Plus at $20 a month. I would not buy a paid seat merely to make occasional checking feel more professional. Start with the access the job actually requires.

Why one good answer is not a trend

The same buyer question can produce different answers across phrasing, days, and engines. That makes a single flattering screenshot a poor basis for a monthly performance claim.

An analysis of 10,826 daily series found that measuring with 20 different prompts was three times more accurate than repeating one prompt over 20 days. The same work reports that 70% of brands appeared on fewer than one in five days, while only 6% appeared on four or more of five days.

The practical distinction is simple: a spot check answers what happened once; repeated checks support a trend. If I only need to know whether ChatGPT can name a brand today, I stay manual and write it down. If I need to say visibility rose or fell this month, I first count the checks required to support that claim.

The hidden manual bill: repetition and interpretation

The 192-check example is deliberately easy to inspect. Twelve prompts are not an enormous research project. Four rounds are not constant surveillance. Yet even at two minutes per answer, collection takes 6.4 hours.

That is where the free route stops being free in any useful accounting sense. Someone still has to open the chats, preserve the wording, save the answers, record citations, and make the next round comparable with the last. Copying and pasting remains work even when no vendor sends an invoice for it.

Then comes the part the collection estimate leaves out:

  • Compare mentions, citations, and recommendations separately.
  • Check which competitors appear on the same questions.
  • Open the cited pages rather than trusting the count alone.
  • Decide whether a difference deserves action.
  • File the evidence so the next round has something to compare against.

Collection time and interpretation time are separate costs. Automating the first does not make the second disappear.

Vendors can obscure that distinction by putting prompts, seats, or domains in the pricing grid while leaving the work around them off the page. I would rather see an unglamorous estimate with editable assumptions than a cheap headline that quietly borrows a day from someone’s calendar.

For an occasional question, that labor may still be the right purchase. For a recurring report, I would price it before assigning it.

What $19 buys: connected evidence, not finished marketing

groas Unlimited costs $19 a month flat for unlimited teammates, ad accounts, and websites, including the data groas collects around the clock. That includes competitor ads and landing pages, Reddit research, AI answers, and SEO data.

I can connect accounts and a website once, then plug groas into Claude, ChatGPT, Gemini, or another agent. The point is not a prettier screenshot. The collected record becomes something I can question in plain English, rather than another pile of answers to copy into a spreadsheet.

A useful starting request is:

Compare our tracked mentions and citations with competitors. Show the underlying answers and dates.

Then I would inspect what comes back. Which engines are covered? How often are answers collected? Can I open the underlying answer behind a count? Does the tracked prompt set match the questions I actually need answered?

Those details determine whether connected collection replaces my manual program. The illustrative 192-check workload is a way to price repetition, not a promise that any subscription reproduces that exact engine-and-prompt schedule.

The manual spot-check versus groas-in-Claude comparison puts evidence and citation quality at the center of that decision. That is the right test. If a visibility claim cannot be traced to a prompt, assistant, date, answer, and cited URLs, I treat it as a screenshot with a subscription.

Connected-route line itemCostWhat I would verify
groas Unlimited$19 a month flatThe tracked coverage and collection frequency needed for the job
Assistant seat$0 on a suitable free tier, or $20 a month for ChatGPT PlusWhether the tier supports the connection and usage required
Review timeSeparate recurring laborAccess to underlying answers and dates
Writing, approval, and publishingSeparate workWho owns the change and who approves it

Monitoring is not writing. It is not legal approval, and it is not publishing. A $19 subscription can collect evidence; it does not make the human work around a page change cost nothing. I keep those hours on their own line.

Where the flat fee matters

The wider market makes the billing unit worth checking. Paid monitoring elsewhere runs from $29 a month at the entry end to $499 or more at the high end. One published comparison lists Semrush AI Toolkit at $99 a month per domain and Ahrefs Brand Radar at $199 a month.

Those headline prices are not a substitute for checking scope. But per-domain, per-prompt, and per-seat charges explain how a small program can become a large invoice.

groas charges $19 flat for the whole team. The second website, the second client, and the extra teammate do not restart the meter. For an agency, that is the whole ballgame. For a single brand, it keeps the recurring software bill clear while the monitoring work grows.

The decision still comes back to labor: how much repeated collection does the connected record actually remove? That is the saving I would count, not every hour anyone spends thinking about visibility.

Spend the next hour on one answerable gap

A record of 192 answers earns its keep when it changes a decision. I would open the collected citations and ask Claude to identify one page gap grounded in them:

  • Which buyer prompt fails to cite the brand?
  • Which competitor page is cited instead?
  • What fact does that page answer that the brand’s page does not?

I will not pay for a dashboard that ends in a report meeting. I want a list I can act on: add the missing specification, fix the conflicting brand fact, or publish the comparison table buyers actually ask for.

Evidence collection and permission to act are different things. I would approve publication explicitly and leave website autopilot off unless I deliberately enabled it. Before enabling anything that can publish live, I would inspect the log.

groas supports that second step too: changes are logged with their reasons and can be undone in one step. That is a control over actions, not a substitute for checking the evidence behind a recommendation.

For the manual route, a tracker that measures mention rates rather than screenshots provides a way to keep the evidence together. Either way, I want the next hour to produce a page change with a date on it, not a larger set of charts.

A check I cannot act on is a hobby. A check that identifies one page to fix is a budget line with a job.

Which bill would I choose?

I would stay manual when the question is occasional and narrow: a product launch, a competitor rumor, or a check that ChatGPT can name the brand today. Run the buyer prompts, preserve the five fields, and stop. There is no prize for turning a one-off question into a recurring subscription.

I would connect the record when repetition becomes the expensive part. If the job requires the same questions across engines and repeated rounds, the illustrative manual bill is 6.4 to 16 hours a month before interpretation. At that point, $19 flat deserves comparison with the collection labor it can replace, with assistant access and review time kept separate.

Neither route buys a guaranteed citation. Free checks do not promise an appearance, and paid tracking does not promise one either. Both buy evidence about where the brand stands. The page fix that follows is still the work.

The cheapest mistake is running free checks with no log, so each answer evaporates and the same curiosity returns next month. The most expensive mistake is paying for tracking, $19 or $499, and never shipping the one page change it pointed to.

Price the questions first, then the repetition, then the fix. Occasional curiosity stays free. A repeatable record gets connected. A dashboard with no publish date is the bill I would never pay.