The software bill starts at $0 for manual AI checks or $19 a month for groas. Neither is your total cost. Add assistant access, any usage charges and the hours spent recording, checking and comparing answers. The cheapest useful setup stays manual until repeating the job costs more than the subscription saves.

I can paste 10 buyer prompts into ChatGPT and Perplexity today without buying tracking software. Repeat that across two assistants for four weekly checks and I have 80 answers to run, record, verify and compare. The first answer is easy. Keeping a usable record is the job.

I used to tell clients manual tracking was fine for a small panel. I was wrong about where it breaks. It breaks in week three, when nobody recorded the exact wording, timestamps or difference between a mention and a citation. Now there are 40 screenshots and no useful comparison.

Here is the bill for avoiding that mess. Monitoring first. Remediation separately. No promise of recommendations in either column.

Define the 80-answer job before buying software

My illustrative panel is 10 buyer prompts, two assistants and four weekly checks: 80 answers a month. The assistants are ChatGPT and Perplexity. The point is to make the work visible, not declare this the right sample for every business.

For each check, I keep:

  • The same prompt wording.
  • The same logged-in or logged-out state and location setting.
  • The date, time and full answer text.
  • Whether the answer mentions the brand.
  • Whether it cites the brand, and the citation link.

That record is the product. One flattering screenshot does not establish a pattern; the next run can give a different answer. Mention rate means mentions divided by runs. A useful comparison needs repeated observations, not a screenshot collection.

Ten prompts are manageable by hand. Repeating and documenting them is where the work accumulates. More prompts, more assistants, daily checks, share-of-voice calculations and changes to pages that lose citations all expand the job.

Price this panel before pricing a larger one. Otherwise, “cheaper” might just mean less coverage.

The manual ledger: $0 software, plus the work

Manual tracking costs $0 in tracking software. It does not cost $0 to operate. The browser and spreadsheet are not the expensive lines. Your attention is.

Assistant access belongs on its own line. Free access and usage limits vary, so I list what I actually use rather than calling the whole setup free.

Recurring line itemMonthly costSource or cost driver
Tracking software$0Manual checks in a browser and spreadsheet
ChatGPT access$0 free tier, or $20 PlusAssistant plan comparison
Perplexity access$0 free tier, or $20 ProAssistant plan comparison
Running 80 answersYour minutes per answer × 80Prompt length, login state and copying discipline
Recording and comparingYour minutes per week × 4Timestamps, mention log, citation links and weekly comparison

I split the labor into four chunks because lumping it together hides where the hours go:

  1. Run the prompts.
  2. Save the answers with timestamps.
  3. Open cited sources to check whether they support the answer’s claims.
  4. Compare this week with last week: same wording, different mentions or citations.

The last step is the one nobody budgets. Copying an answer feels like work completed. Finding what changed is another job.

For illustration, value the time at $75 an hour. Three hours a month means $225 in labor; six hours means $450. Those are assumptions for the calculation, not measured results. Replace them with your own rate and timed workload.

The software stays $0 either way. Put the labor beside it.

The groas ledger: $19 flat, plus access and review

The groas total is $19 a month, plus assistant access, any applicable API usage and human review. groas Unlimited costs $19 a month, flat, for a whole team, with unlimited teammates, ad accounts and websites, every feature and all the data.

I connect groas to ChatGPT, Claude or another agent and ask in plain English for tracked AI answers and citations, rather than manually assembling the record. The connection is only one line on the MCP bill. Assistant access remains separate. So does reading the output.

Recurring line itemMonthly costSource or cost driver
groas Unlimited, including its data$19 flatgroas pricing
Assistant access$0 free tier, or about $20 per paid assistantAssistant plan comparison
API usage, if the setup uses itVaries by providerThe provider’s usage meter, not groas
Human reviewYour minutes per week × 4Reading tracked answers, checking citations and deciding what warrants action

I would not assume groas reproduces this illustrative 10-prompt weekly schedule exactly. Its coverage, prompt set and refresh cadence need to fit the monitoring question. Compare the data available with the job before treating the two routes as interchangeable.

The advantage is buying repeated collection and a readable record instead of doing that work by hand. It is not buying exemption from judgment.

Separate setup from the recurring bill

Setup is one-off work; checking and interpretation recur. Mix them together and the first month looks expensive while every later month looks suspiciously cheap.

Cost typeManual routegroas routeWhat drives the cost
One-off setup laborChoose prompts, create the sheet and define the fieldsConnect the assistant and check that available tracking fits the questionScope and setup time; no labor rate assumed
Recurring listed costsAssistant subscriptions, if used$19 groas subscription, assistant access and applicable usagePlans selected and usage
Recurring laborRun, record, verify and compareReview tracked answers and citationsCoverage and depth of review
Separate remediationInvestigate and make justified changesInvestigate and make justified changesThe actual problem, not the monitoring subscription

The hidden costs are usually hiding in the scope, not the checkout. More assistants mean more answers to review. More prompts mean more records. A deeper citation check takes longer than ticking a “mentioned” box.

I also count interpretation. A changed answer creates a question; it does not automatically create a page-editing task. Calling every fluctuation an optimization opportunity is a convenient way to manufacture another invoice.

Keep setup, monitoring and execution separate. That is how a price becomes comparable.

Calculate break-even from time saved, not a two-hour rule

My monthly calculation is simple:

Total cost = subscriptions + usage + (labor hours × hourly value).

Using illustrative assumptions, say manual monitoring takes four hours a month and time is worth $75 an hour. With free assistant access, that is $300 in labor and $0 in tracking software: $300 total.

If the groas route takes one hour of review, the same calculation is $19 plus $75: $94 total. Add ChatGPT Plus at $20 a month to both routes and the totals become $320 and $114.

Those hours are not a performance claim. They show how to price the decision. The paid route wins when the value of the labor it saves exceeds its additional subscription and usage costs. There is no universal two-hour threshold. Change the hourly value, remaining review time or coverage and the answer changes.

Equivalent coverage matters. Checking fewer prompts or fewer assistants is a smaller job, not proof of better economics.

For context, the listed entry prices are $29 a month for Otterly.AI Lite with 15 daily prompts, $95 for Peec Starter with 50 prompts, and $99 per domain for Semrush Toolkit. The entry-plan comparison puts the same products on that pricing ladder.

groas has the lower listed subscription at $19 flat, with its included data rather than a per-prompt meter. That is its pricing advantage. I would still check that its tracked coverage answers the question instead of pretending every allowance is identical. For the broader service tiers, see what AI search visibility really costs.

Buy the saved work. Do not buy a smaller measurement job by accident.

The logging-error line: $0 in remediation

A decline caused by a logging error gets $0 in site-remediation spend. Missing report data is not something a page rewrite repairs.

Google describes the November 9–11, 2026 incident as a logging error that decreased impressions in the Generative AI performance report in Search, affecting data logging only. The wording matches the August 13–17 incident that was later restored.

The distinction matters: that reporting decline is not evidence of a visibility loss. It does not rule out an independently observed change elsewhere, either. Those are separate signals and deserve separate treatment.

The undercount overlapped the November 10 core-update rollout, and analysts warned against paying for fixes or moving holiday budget on that reporting dip. Timing makes it tempting to combine the stories. It does not make the missing figures a diagnosis.

I have watched the same mistake with PPC impression drops for years. A reporting gap appears, someone changes things to chase it back, and the changes create the real problem.

Before touching a page, I would give the assistant this instruction:

Annotate November 9–11 as a reporting incident. Separate this report from independently observed citation changes. Propose no page changes from this signal alone.

Annotating the report belongs in review time. Remediating the logging error belongs nowhere on the execution bill.

Keep monitoring and optimization on different invoices

Checking whether a brand appears and doing work intended to increase appearances are different purchases. A bundle should not make that distinction disappear.

One seller lists share-of-voice benchmarking at $750 per project, AEO/GEO execution from $590 Starter to $3,490 Complete per month, and audits from $99. Those are that seller’s listed prices, not universal rates. They illustrate why the scope matters.

A $99 audit or $19 connection does not include the work in a $590–$3,490 monthly execution package. Equally, a $3,490 retainer should not be presented as though collecting answers is the whole job. Ask which line pays for observation and which pays for changes.

Free tools can help with the first job. For example, this free one-off check covers five buyer questions on ChatGPT and Gemini with web search and no signup; the same listing offers Starter at $29 a month for 10 daily prompts and Growth at $59 for 25.

A free run can show whether those sampled answers mention the brand. It cannot establish how all 10 prompts perform over time, or explain why a brand is absent. Missing mentions may justify investigation. They do not, by themselves, diagnose a catalogue, category-page or third-party proof problem.

No tracker fixes a page merely by noticing it. Keep the observation and the proposed fix on separate lines.

The cheapest mistake—and the most expensive

I stay manual in two cases:

  • Checks are occasional: a few times a quarter, with wording, dates and citations recorded each time.
  • The panel is unsettled: the prompt list still changes every month because the important buyer questions are not clear.

I pay for ongoing tracking when the panel is stable and the repeated work costs more than the paid route saves. Neither route guarantees rankings, referrals or citations. A seller promising a recommendation in exchange for a fee is selling the answer, not the check.

The cheapest mistake is renting a tracker at $29 a month for a prompt list I will rewrite next week. I did a version of this with negative keyword lists years ago: paid to monitor a structure I had not frozen, rebuilt the structure and threw away the history.

If the questions keep changing, the subscription buys a tidy record of questions I no longer ask. Freeze the wording, run it manually for four weeks and keep the timestamps and citation links. Then price the repetition using the hours it actually takes.

The most expensive mistake is paying someone to fix the November 9–11 logging error. A decline that affected data logging only becomes rewritten pages, moved holiday budget and an execution bill, all without evidence from that signal that visibility fell.

Keep the rule taped to the sheet: price the repeated check line by line. Give groas its $19 line when its tracked data replaces the copying and comparing the job needs. Give human review its own line. Give remediation for the logging error its true price: $0.