How to track your brand in AI search

Tracking your brand in AI search is a repeat test: you put the same buyer questions to ChatGPT, Perplexity, Google's AI answers and other engines on a schedule, then log whether your brand is named, where it sits in the answer and which sites the answer cites. We follow 48 AI visibility trackers, and the numbers below come from that dataset, with prices checked Oct 17–27, 2026. 43 of the 45 name ChatGPT, 17 re-run prompts daily on their entry plan, and the median entry plan costs $1.57 per tracked prompt a month across the 28 tools that publish a USD price and a prompt count.

The five steps

1. Write a fixed prompt list. 20 to 50 questions your buyers ask before they know your name. 2. Choose engines. Start with the four that most trackers carry, then add the rest only where your buyers are. 3. Set a re-run schedule. Weekly for a monthly report, daily when you're testing a change. 4. Work out the budget. Prompts × price per prompt, with the schedule as a cross-check. 5. Read trends per engine. One answer is a single sample; a month of re-runs is evidence.

Start by hand

You can run a baseline without a tool. Open ChatGPT and Perplexity in a logged-out window, ask each of ten buyer questions, and write one row per answer in a spreadsheet: date, engine, prompt, brand named (yes or no), position in any list, competitors named, and the domains cited. Paste each answer's cited links into a column too, so you can tally which domains recur. Repeat a week later with identical wording. Two passes reveal how much answers drift before you pay anyone to automate it.

Our AI Visibility Check is GEO Tool Index's own product. It asks ChatGPT one buyer question for free; the $19 report runs 15 prompts on ChatGPT and Perplexity, 30 dated answers in all. It is a one-time snapshot, not a tracker, and it isn't listed in our tool tables.

A paid tracker isn't always worth it yet. If your market is a narrow niche with a long sales cycle, a quarterly hand check of a dozen prompts may tell you enough. The same goes for a seasonal business outside its season, or a team with no capacity to act on findings for months. Automate once you have a launch, a content push or a competitor you need to watch week to week.

Step 1: write the prompts

The median entry plan in our data allows 50 prompts, and 18 of the 28 priced entry plans stop at 50 or fewer, so a budget tracker holds one product line, not a whole catalogue. Spend the allowance on four kinds of question:

• Category: "best payroll software for restaurants". This shows whether engines recommend you unprompted.
• Comparison: "[your brand] vs [competitor] for a 20-person team". Here the order of the brands in the answer matters.
• Problem: "how do I run payroll for tipped staff". Look at which pages the answer cites; that is whose content gets used.
• Brand check: "is [your brand] any good". This shows what the engines say about you when asked directly.

Sample sets for three kinds of business, written the way a buyer types:

A local business needs the town or neighbourhood in the prompt, because engines answer local questions for a place. An agency should keep one list per client rather than one shared pool, so each client's trend stays separate.

Keep the wording frozen once tracking starts. A reworded prompt is a new prompt, and its history starts again at zero.

Vendors count a "prompt" differently, which changes what an allowance buys:

• Peec AI Starter: "50 prompts, run daily, 3 models". The prompt allowance and the model count are listed separately.
• Searchable Pro: "100 prompts per model", so three engines use up the allowance three times faster.
• Trakkr Growth: "50 prompts per brand".
• Ahrefs Brand Radar sells checks, and its notes define one check as one prompt on one platform.
• Wellows Basic sells 1,800 AI answers a month instead of prompts; Rankscale and AthenaHQ sell credits.

Step 2: choose engines

Each tracker names its engines on its pricing page. Across the 48, this is how many name each one (name variants such as "ChatGPT Search" count under ChatGPT):

28 of the 45 name all four of ChatGPT, Perplexity, Gemini and Google AI Overviews. The median tracker names 6 engines and the widest names 12. AirOps names ChatGPT alone. Bluefish and Brandlight name no engine on the pages we read, so their coverage is unknown.

A name on the pricing page doesn't put that engine on the cheapest plan. AEO Vision lists 6 engines, yet its Lite plan tracks ChatGPT only. BrandAxis paid plans list 3 models, and Superlines Starter covers 3 engines. Check the plan card, not the logo strip.

If you don't know where your buyers ask, start with the four most trackers carry. Add Google AI Mode for search-led categories, and Claude or Copilot only when your buyers work in those tools.

Step 3: set the re-run schedule

AI answers change between runs, so a tracker repeats each prompt on a schedule. On the entry plan, the cheapest listed price, the 48 trackers state these schedules:

Daily on the entry plan: BrandAxis, CueScout, Gauge, Geoptie, Goodie AI, Nightwatch, Otterly.AI, Peec AI, Rankability, Rankshift, SE Visible, Searchable, Semrush AI Visibility Toolkit, Superlines, Trakkr, Writesonic and ZipTie. The 6 weekly ones include LLMrefs and Wrenda. AEO Vision and Promptmonitor run daily from their second plan up (Promptmonitor's page also says every plan includes daily tracking, while its Starter card says twice a week, which is what we count). Wrenda moves to daily on its Scale plan. The 16 that don't state a schedule include the 4 trackers with no public price.

The schedule multiplies your data. Checks a month = prompts × engines × runs. 25 prompts on 3 engines come to 2,250 checks a month run daily, against about 321 run weekly. 50 prompts on 4 engines daily is 6,000.

Weekly runs suit a monthly client report. Daily runs earn their cost in the weeks after you publish or change a page, when you want to see an answer shift.

Step 4: set the budget

The median entry plan costs $1.57 per tracked prompt a month (28 tools). Prices run from $0.16 a prompt on LLMrefs All in One to $3.96 on Semrush AI Visibility Toolkit Base, and 10 of the 28 cost $1 or less a prompt.

At the median rate, our arithmetic gives:

The cheapest plan we track with 100 or more prompts for one brand is SearchInsight Starter at $59 a month, billed annually. Across the 38 trackers with a USD entry price, the median entry plan is $74 a month.

Price per prompt ignores the schedule. LLMrefs All in One costs $79 for 500 prompts run weekly; SE Visible Basic costs $89 for 75 prompts run daily. Spread over a month of re-runs, Rankshift Starter is the cheapest at $0.020 a run and Wrenda Starter the dearest at $0.695. The median across the 24 trackers that state a schedule is $0.066 a run.

The full table is on cost per prompt. For a capped budget, see AI visibility tools under $50 a month; for ChatGPT alone, our ChatGPT tracker ranking.

Mistakes that skew the numbers

• Tracking from your own account. Chat history, memory and custom instructions can colour what ChatGPT tells you. Use a fresh session, or a tracker that queries without your login.
• Ignoring location. Ahrefs Brand Radar defines a check as one prompt on one platform in one location, which is a reminder that a city or country setting is part of the prompt.
• Pooling branded and unbranded prompts. A question that contains your name tends to return it, so a pooled mention rate looks healthier than it is. Report the two groups apart.
• Comparing two vendors' scores. Each tool samples, counts and weights differently. Compare a tool with itself over time, not with a rival's dashboard.
• Reacting to one drop. A brand that vanishes from one answer can be back the next run. Wait for the pattern to hold across several runs before rewriting pages.

Step 5: read the results

Read each engine on its own line. The same prompt can name you in Perplexity and leave you out of ChatGPT, and an average across engines hides that gap.

Four readings carry most of the signal:

• Mention rate: the share of answers that name your brand, per engine, per week.
• Position: where you appear when an answer lists several brands.
• Who appears instead: the competitors named in answers that leave you out.
• Cited sources: the domains an answer links to. Pages that keep getting cited are where buyers' questions get answered.

Treat one week as noise and four weeks as a trend. Our own check page warns that "Answers can vary by date, location and account", and that "API answers approximate consumer apps but are not identical." Trackers differ on that second point, so ask a vendor whether it reads the consumer app or the API before you compare its numbers with another tool's.

Questions to put to a vendor

Pricing pages leave a lot out. Before a trial ends, ask:

• Does one prompt cover every engine, or count once per engine? Step 1 shows both models in use.
• Does it query the consumer app or an API? Results from the two can differ, so the answer explains gaps between tools.
• Can you export raw answers and citations? A CSV or API export lets you audit a score; a dashboard-only plan doesn't.
• How long is history kept, and does it survive a downgrade? Our dataset doesn't record history retention for any of the 45, so treat it as unknown until the vendor confirms in writing.
• Are personas or locations extra? Wrenda Starter allows up to 2 personas and meters a quota of prompt × persona × platform.
• Does the schedule change after onboarding? Scrunch Core refreshes daily for the first 14 days, then every 3 days.

A first month, week by week

• Week one, baseline. Load the frozen list, tag every prompt as branded or unbranded and by buying stage (researching, comparing, deciding), and note each engine's mention rate. That snapshot is your benchmark.
• Week two, gaps. Sort the unbranded prompts where you're missing. Write down which rivals fill those slots and copy the cited URLs into a shortlist.
• Week three, targets. From the shortlist, pick a few cited pages where a mention is realistic: a directory entry, a comparison article, a forum thread, a review roundup. Then check that your own site answers at least one problem question plainly, in its first paragraph.
• Week four, prune. Retire prompts that return no brand at all, swap in phrasings customers use (sales calls, support tickets and site search are good mines), and judge whether daily runs earned their price.

Terms you'll meet in dashboards

• Mention: your brand's name appears in the answer text.
• Citation: the answer links to a URL, yours or someone else's, as a footnote or source card.
• Share of voice: your mentions divided by all brand mentions across a prompt set. Vendors define the denominator differently, so the figure travels badly between tools.
• Sentiment: a label (positive, neutral, negative) that a model assigns to how you're described. It's an automated judgement; spot-read the underlying wording before quoting it.
• Visibility score: a vendor's own blend of the above. Ask for the formula, or rely on the raw counts.

Reporting it to your team

A one-page monthly note keeps tracking from turning into a dashboard nobody opens. Lead with one sentence on direction (up, flat or down per engine), then three short blocks: the rivals gaining ground, citations won or lost, and prompts added or retired. Close with the two or three actions for next month and who owns each. Marketing, SEO and PR all touch the outcome, so name an owner per action rather than per channel.

FAQ

What does AI search tracking cost?

The median entry plan is $1.57 per tracked prompt a month across 28 trackers, so 50 prompts come to about $78.50 at that rate. The cheapest per prompt is LLMrefs All in One at $0.16, run weekly.

How often should prompts re-run?

17 of the 45 trackers re-run daily on their entry plan and 6 weekly; 16 don't say. Weekly fits a monthly report, and daily helps while you test a change.

Can I track AI visibility for free?

By hand, yes: a spreadsheet and two passes a week apart. 4 of the 45 trackers keep a free plan: AthenaHQ, BrandAxis, CiteCycle and Qwairy.

Our counting rules

Counts come from our dataset of 48 trackers, with prices checked Oct 17–27, 2026 (newest check Oct 27). Engines are the ones each vendor names on its pricing page. The schedule is read from the entry plan's prompt allowance text, or from its plan notes where the allowance is blank. Wellows' "≈60/day" is an answer quota, not a schedule, so we count it as not stated. Cost per prompt is the entry plan price divided by its prompt allowance; our methodology sets out the source rules.

We have affiliate links with AirOps, LLMrefs, Nightwatch, Peekaboo, Rankscale, Rankshift, SE Visible, Searchable, Wellows and Writesonic, all named on this page, so a purchase through our links can earn us a commission. No count, price or list on this page reads affiliate status.

Featured listings, sponsor slots and affiliate links are labelled, and none of them changes the data, prices or the order of the data tables.