A B2B company can track whether AI assistants recommend it with a free manual prompt log: 20 to 30 buyer questions, asked monthly in fresh chats on the main assistants, recorded as named, cited or absent. Paid tools start around $29 a month and automate the same repeat checks; they add speed and history, not magic.
The log comes first even if you plan to buy, because it teaches you which questions matter and what a good answer looks like. A tool run on poorly chosen prompts simply measures the wrong thing at scale. The sections below set out the log, the cost bands, and the few numbers worth watching.
Why the answers move and what that means for measuring
Search engines return a ranked list that changes slowly. An assistant composes an answer, and the composition varies between runs, between accounts and between days. Ask the same question five times and you may see your company named in three. That is not a fault in your measurement; it is how the product behaves. So the unit to track is a rate over several runs, such as "named in 3 of 5 runs", not a single yes or no.
A second feature: answers depend on whether the assistant searches the web at the time. Some answers come from what the model learned in training, others from live search results with links. Record which one you saw, because they respond to different work. Being cited from live search responds to the pages you publish and the sites that mention you. Being named from memory changes slowly and mostly follows how widely you are discussed.
The free prompt log, built in two hours
Choose the questions
Write questions in the words a buyer would type, not in your marketing language. Group them in four buckets: problem questions ("how do we reduce invoice errors in a ten-person finance team"), category questions ("best accounts payable software for small manufacturers"), comparison questions ("X versus Y for a company our size") and trust questions ("is [your company name] any good", "[your company] pricing"). Take real wording from sales calls, support tickets and the search terms in your analytics. Aim for 20 to 30 questions; five per bucket is a fair start.
If you want help drafting them, give an assistant your product description and ideal customer and ask for 40 questions buyers ask at each stage, then trim to the ones you recognise from real conversations. The recognition step matters, because invented questions measure invented demand.
The Complete Excel Course with AI Integration
From formulas and pivot tables to Power Query, DAX, VBA and AI with Copilot, ChatGPT and Claude — hands-on from the first lesson.
By Sagnik Bhattacharya and Virginia Thorn, Primetime Emmy Award winner
Buy nowSet up the sheet
One row per question per assistant per run. Columns: date, assistant, question, run number, were you named (yes or no), position (first, in a list, last), were you cited with a link, which competitors were named, was anything said about you wrong, the answer pasted in full, and the source links shown. Keep the full answer, because the notes you write later are only as good as the evidence behind them.
Run it cleanly
Use a fresh chat for every run, logged out or with memory and personalisation switched off where the product allows it, because earlier conversations colour answers. Ask each question three times on each assistant. Check the assistants your buyers actually use; for many B2B buyers that means ChatGPT, Google's AI Overviews and AI Mode, Gemini, Microsoft Copilot and Perplexity, but ask your last ten customers which they use rather than assuming. Thirty questions, three runs, three assistants is 270 checks, which takes a couple of hours with a stopwatch and a spreadsheet, so be selective in the first month.
An illustrative log and what it shows
Say a twelve-person company sells scheduling software to equipment-hire firms. It logs 25 questions, 3 runs each, on 3 assistants: 225 checks. Illustrative results below.
| Bucket | Checks | Named | Cited with link | Rate named |
|---|---|---|---|---|
| Problem questions | 45 | 4 | 2 | 9% |
| Category questions | 45 | 11 | 6 | 24% |
| Comparison questions | 45 | 18 | 9 | 40% |
| Trust questions (own name) | 45 | 40 | 31 | 89% |
| Pricing and alternatives | 45 | 9 | 3 | 20% |
The pattern is typical: the company is well known by its own name but nearly invisible when buyers describe a problem. The useful reading is not "we are at 9%" but "buyers who do not yet know us never meet us". That points to content and third-party mentions answering problem questions, which is work, not a tool purchase. The log also shows competitor frequency: say one rival is named in 31 of 45 category checks. That tells you whose pages and review profiles the assistants lean on, and which sites to study.
A sample entry and what to fix in it
Date: 3 Nov | Assistant: ChatGPT (search on) | Q: best scheduling software for small equipment hire firms | Run 2
Named: no | Competitors named: Rival A, Rival B, Rival D | Sources shown: rival page, two review sites, one forum thread
Notes: review site 1 lists Rival A, B, D but not us; forum thread is two years old and recommends A.The action in that row is not "optimise for ChatGPT". It is "get listed on review site 1 and answer the forum thread with a factual reply". Notes should always end in something a person can do.
Cost bands for paid tools
Paid trackers run your prompts on a schedule, store the answers and chart the trend. Prices below are list prices read from each vendor's own page in November 2026, on monthly billing where the page offers a choice; they are changing quickly across the whole category, so check before buying.
| Band | Example | List price | What the price buys |
|---|---|---|---|
| Entry | Otterly Lite | $29 a month | 15 search prompts checked daily; tracks ChatGPT, Google AI Overviews, Perplexity and Copilot, with Claude, AI Mode and Gemini as add-ons |
| If you already use Ahrefs | Ahrefs custom prompts | Included checks on paid plans (Lite is $129 a month); extra packages from $50 a month | Lite includes 150 checks a month, about 5 prompts daily on one platform (Standard 300, Advanced 600); the $50 package adds 2,500 checks |
| Small | Peec AI Starter | $95 a month | 50 prompts, one country, ChatGPT, AI Mode, AI Overviews, Copilot, Gemini and others, three active at a time |
| Small to mid | Otterly Standard | $189 a month | 100 search prompts, extra 100 at $99 |
| Mid | Peec Pro; Otterly Premium | $245; $489 a month | 150 and 400 prompts respectively; Peec Pro covers three countries |
| Index-style | Ahrefs Brand Radar AI Visibility Index | From $199 a month | Draws on a database its page puts at 531 million prompts rather than only yours |
Three cautions on that table. Otterly offers 15% off for annual billing, so confirm at checkout which price you are paying. Ahrefs counts checks rather than prompts: one prompt on one platform in one location, checked once, is one check, so a prompt run daily on two platforms uses about 60 checks a month, and its page bills overage at the end of the month. The Brand Radar page also shows a "$699 a month for all models" figure without saying which tier it covers, so ask before assuming. None of this is a criticism; it is a reminder that pricing in this category moves quickly, so a comparison article older than a few months may be out of date. Check each vendor's page on the day you buy.
For agency-side detail on the two big suites, see Semrush AI Visibility vs Ahrefs Brand Radar.
What 15 prompts actually covers
A cheap plan with 15 prompts sounds limiting, and it is, but 15 well-chosen prompts can show the direction. Consider the sums. At 15 prompts across four assistants you have 60 prompt-engine pairs. Otterly checks daily, so the tool collects about 1,800 answers a month, plenty to estimate a rate. Peec's page does the same sum for its Starter plan: 50 prompts on three models, checked daily, is shown as 4,500 answers a month. The limit is breadth: 15 prompts cannot cover problem, category, comparison and trust questions in each of your three market segments. So choose the prompts that matter most to revenue: the comparison and category questions your best customers asked before buying.
Example: choosing a plan by prompt budget
The scheduling-software company above has 25 questions. Option A: free log for three months, then a $29 plan on the best 15. Option B: a $95 plan with 50 prompts straight away. If the owner will read the reports only once a month, the free log is enough for now, and a $95 plan would mostly generate charts nobody opens. If the owner has an agency or an in-house marketer who will act weekly, the $95 plan with 50 prompts, one market and the assistants buyers use is better value. The deciding factor is who acts on the data, not how much of it exists.
Five things to measure and the ones to ignore
- Named rate on category and comparison questions. The number of checks in which you appear, out of all checks, per bucket.
- Cited rate. Checks where the answer links to your own pages.
- Accuracy. Count answers that state something wrong about you, such as an old price or a discontinued product. Wrong facts spread from stale pages, and fixing the source page is the remedy.
- Competitor share. Which rivals appear beside you and what sources support them.
- Referral traffic. Visits from assistants in your analytics, which tie visibility to enquiries.
Ignore composite scores you cannot trace back to specific answers. Ask any vendor whether you can open the underlying answers behind a number. Also treat sentiment scores with care: a score is a model's summary of a model's answer, so read ten samples before relying on it.
Connecting visibility to enquiries
Visibility is only useful if it leads somewhere. Add a question to your enquiry form or sales call: "How did you first hear about us?" with "an AI assistant" as an option and a free text box. In analytics, group referral visits from assistants into one channel. A B2B company with 40 enquiries a month might see 3 mention an assistant. That is small but informative, because those buyers arrive having already seen a recommendation. Keep a note of what they asked, since it will often be wording missing from your prompt log.
Letting an assistant summarise your log
Once the sheet has a few hundred rows, an assistant can do the first read. Export the sheet as a CSV, paste it in sections if needed, and give a prompt that forbids speculation.
Below is a log of AI-assistant answers to buyer questions for [company]. Columns: date, assistant, question, run, named, position, cited, competitors named, errors, sources.
1. Give the named rate and cited rate per question bucket.
2. List the five competitors named most often and the sources most often shown beside them.
3. List every row where "errors" is not empty, with the exact wrong statement.
4. Say which questions had the biggest gap between named rate for competitors and for us.
Report only what the rows show. If a figure cannot be computed from the data, say so.
DATA:
[paste CSV]Illustrative reply: "Category bucket: named 11 of 45 (24%), cited 6 of 45 (13%). Rival A named in 31 of 45 category checks, with a review site and a trade publication shown beside it in 22 of them. Two rows record an old price of $49 a user; the current price is $59." Spot-check three of those figures against the sheet before acting, because assistants mis-add. The value is in the shortlist: two sources to approach, one wrong price to correct.
A one-page monthly report
Keep the report short enough that someone reads it. A workable layout has five lines: the headline rate on comparison questions against last month; the bucket that moved most; the one wrong fact found and where it came from; the competitor and source gaining ground; and the single action chosen for the coming month, with an owner and a date. An example: "Comparison named rate 40%, up from 33%. Problem questions flat at 9%. Wrong price on a directory page, update requested 5 Nov. Rival A cited via a review site we are not on. Action: apply for the listing, owner [first name], by 20 Nov." One action a month, finished, beats a dashboard of ten.
Example: a B2B manufacturer with a thin website
A fifteen-person company making custom steel brackets sells to other manufacturers. Its comparison and trust questions look fine, but when it logs problem questions such as "how do I source custom brackets in small batches" it appears in none of 45 checks. The sources shown are two trade directories and a supplier-guide page on a rival's site. The company's own site has a product catalogue and no explanatory pages. It spends a month publishing three plain pages: how small-batch bracket orders work with lead times, a drawing checklist for quoting, and a materials comparison with real price ranges from its own quotes. It also applies to both directories. Two months later the log shows named in 5 of 45 problem checks and cited in 3. That is a modest rise, but it came from content and listings that cost no software.
Example: the wrong answer you did not expect
A software company finds that three assistants state it has no free trial, because an old comparison page says so. The company has had a trial for a year. The fix is not in the assistant: update the comparison page's owner, correct its own pricing page so the trial is stated in plain text near the top, and add the fact to its FAQ. Re-run those three prompts a fortnight later. If answers that search the web now say the trial exists, the remaining wrong answers come from training data and will correct slowly, so log them and move on rather than spending money chasing them.
When a paid tool is clearly worth it
Move from the log to a tool when the manual run takes more than three hours a month, when more than one person needs the numbers, or when you need daily history to link a change on your site to a change in answers. A marketer who spends a whole afternoon each month copying answers into a sheet costs more than a $95 plan, so the tool pays on time alone. Keep the manual log for a handful of questions even then, as a sanity check against the tool's own numbers.
Sampling rules so the numbers hold up
Keep the same questions each month, with the same wording, and only change the set on a fixed date, recording the change so a jump in the rate is not caused by an easier question. Do not run the prompts while logged in to an account with a long history on that assistant unless that is how your buyers use it. Do not cherry-pick: run all three attempts and count all of them. If you use a paid tool, check that its prompt wording matches yours, because small wording changes shift results. Finally, compare yourself with two or three named competitors using identical prompts, since an absolute rate means little until you see who else is named.
Fixing what the log reveals
Three kinds of fix recur. First, absent from comparison pages: the assistants quote third-party lists, so ask to be included in relevant roundups and keep your listing on review sites accurate. Second, wrong facts: update your own pricing, product and about pages, and any directory entries that repeat old information, because stale pages feed stale answers. Third, thin pages: buyers ask problem questions and your site answers with product features. Publish a page per real buyer question, with specific numbers and an honest comparison, as in the tutorials on getting a practice recommended in ChatGPT search and AI search for a trade business.
Re-run the log a month after each change, on the same questions, so you compare like with like. Expect slow movement: a new page can be cited within days where the assistant searches live, and take much longer where it relies on training data.
Questions for any vendor before you pay
- Which assistants are included in the price, and which are add-ons?
- How many runs per prompt per day, and can I see every raw answer?
- Is the price per prompt, per check or per brand, and what happens when I go over?
- Can I track my own prompts, or only a vendor-supplied set?
- Can I export the data if I leave?
- What does annual billing change about the price shown?
More questions of this kind are collected in what a small firm should ask an AI visibility provider. Take the cheapest plan that covers your top 15 prompts on the assistants your buyers use, give it three months, and cancel if nobody has acted on a report. If you would like someone to help choose the questions and read the first results, that is the kind of work my AI implementation consultation covers.
Questions about tracking AI recommendations
Can I track AI visibility for free?
Yes, at small scale. Write 20 buyer questions, ask each in a fresh chat on the assistants your buyers use, and log whether you are named, cited and how. Repeat monthly. It takes about two hours a month and is enough to see direction.
How many prompts should a B2B company track?
Start with 20 to 30 questions covering what buyers ask when comparing vendors, solving the problem and checking pricing. More prompts add cost before they add insight. Expand only after you act on what the first set shows.
Why do results differ each time I ask?
Assistants generate answers afresh, so wording, order and the brands named vary between runs. Treat one answer as one sample. Run each prompt a few times and report a rate, such as named in 3 of 5 runs.
Is a share-of-voice score reliable?
Only as reliable as the prompts and the number of runs behind it. Check what the score counts, which prompts feed it and whether you can see the underlying answers. A score you cannot trace to answers is a vanity number.
Further reads
- AI Visibility for Hotels: Get Recommended in ChatGPT Answers — A sector example of visibility work.
- AI for Ecommerce Analytics: Shopify Reports, GA4 or ChatGPT First? — Decide which reports to read first.
- What Are the Best AI Visibility Products for a Local Business? — Visibility products compared for local firms.
- How to Check Sources and Citations in AI Research — Check the sources an assistant cites.
- How Wedding Businesses Get Found When Couples Ask AI for Suppliers — How a small supplier gets named in AI answers.
- How to Monitor Brand Mentions in ChatGPT for a Client, Week by Week — A fixed prompt set, repeat runs, a log sheet and a weekday schedule for tracking a client's ChatGPT mentions, with tool costs and wrong-answer fixes.
- Best AI Brand Visibility Software: Is AI Mentioning Your Business? — Check by hand first. Paid AI visibility trackers start around $29 a month, and only some small businesses need one. Prices, limits and a test.
- Best AI Tools for B2B Marketing at a Small Manufacturer or Distributor — Ten AI tools matched to industrial marketing jobs: case studies from job notes, datasheet copy, trade emails and follow-ups, with prices and examples.
- AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.
Sources: Otterly and Peec AI pricing pages, Ahrefs pricing and Brand Radar pages, read in November 2026.