Manual AEO tracking vs a tracker: the 3-minutes-an-answer math

Checking AI visibility by hand is one calculation: prompts × engines × runs a month, at about three minutes an answer. That is the arithmetic behind the calculator on our pricing page. This post works it through honestly, including the cases where a spreadsheet is the right tool.

What one manual check involves

Open the engine. Turn web search on. Paste the question. Read the answer to the end. Note whether you were named, where in the list, whether your site was cited, who else was named, and which pages the answer was built from. Paste the answer somewhere you can find it next week. Three minutes is a fair estimate for a careful person; the note-taking takes longer than the reading.

Most hand-kept sheets drop the last two fields first. After a few weeks they hold a column of yes and no, which is the one thing you cannot act on. The fix for a lost question is on the pages the engine read, so a sheet without the sources is a scoreboard, not a plan.

The arithmetic

Answers a month = prompts × engines × runs. Hours = answers × 3 ÷ 60. Working days = hours ÷ 8.

Setup	Prompts	Engines	Runs a month	Answers a month	Hours at 3 min	Working days
A founder's spot check	5	2	1	10	0.5	0.1
A monthly spreadsheet	10	5	1	50	2.5	0.3
Weekly, one product	25	5	4	500	25	3.1
Weekly, three sites	75	6	4	1,800	90	11.3
Weekly, ten client sites	250	6	4	6,000	300	37.5
Daily, ten pinned questions	10	6	30	1,800	90	11.3

The rows match the plans. Starter is 25 questions on 5 engines weekly (125 answers a week, $39 a month). Pro is 75 questions on 6 engines over 3 websites (450 a week, $99). Agency is 250 questions on 6 engines over 10 sites (1,500 a week, $249). Enterprise is $799. Pinned questions are checked daily on top: 3 on Starter, 10 on Pro, 25 on Agency.

When a spreadsheet is fine

The first two rows. If you have five questions, care about two engines, and want to know once a month whether anything changed, that is half an hour. Do it by hand. Keep the answers, not just the verdicts, and write the date on each. Three things make the manual version work:

• Use a fresh session with web search on and no chat history, so the answer is what a stranger would get.
• Ask the exact same wording each time. A reworded question is a different question.
• Record who was named instead of you and the source list. The verdict "not named" is the least useful field.

When it stops being fine

The third row is where it breaks: 25 questions on five engines every week is 500 answers a month, or 25 hours. Nobody does 25 hours of pasting every month for long. What actually happens is that the check gets skipped in a busy week, then two, and the series has gaps at exactly the moments something changed. The cost of manual tracking is not the hours. It is the missing weeks.

Three more things the spreadsheet cannot do at that scale:

1. Consistency. The tracker asks every engine the same question with web search on, once, at the same low temperature, and stores the whole answer with its sources. Two people checking by hand get two methods.
2. The range. A score with a ± tells you whether this week's move is real. Hand-counted mentions cannot carry one. The run-to-run flip rate on the platform is 3.4% (September 2026), and you need the sample size to know whether a change beats it.
3. The sources. Across one account's 34 questions the engines cited 1,685 sources over 640 URLs and 380 domains. Counting which domains recur is the step that finds the fix, and it is the step nobody does in a spreadsheet.

What the plans do with the same arithmetic

A plan is a number of answers read for you every week. On Starter that is 125 answers a week and about six hours of manual checking. On Pro it is 450 answers and more than 22 hours. On Agency it is 1,500 answers and roughly 75 hours, which is two analysts' full weeks. Every plan reads each answer the same way: named or not, position, citation, tone, who else was named, and the sources. The first check runs about five minutes after setup. As of 29 September 2026 the platform had read 6,357 answers, the count on our homepage. The methodology page has the formula each answer is scored with.

Other tools price the same units differently. How AI visibility tools price prompts compares per-prompt, per-engine costs across fourteen products, and How to track AEO covers the manual method step by step.

A rule of thumb

Multiply your prompts by your engines by your runs a month. Under 60 answers, use a spreadsheet and be strict about the method. Over 300, that is 15 hours a month of pasting, and the $39 plan reads 500 a month with the sources attached. Between those, decide on whether you need the sources and the range, not on the hours.

Count your own answers

The free audit is one answer done for you: your domain, one question your buyers ask, two engines, and who they named, in about a minute with no account. Then start the trial at Pro capacity, 14 days, no card, and let it read the rest. The Named Report will publish how many answers the platform reads each month, how many cited the business asked about, and how many named it.