DeepSmith

Nov 26 · AEO & AI Visibility

16 min read

How to Track AI Visibility on Under $300 a Month

Avinash Saurabh
Avinash Saurabh · CO-Founder & CEO
Dark monochrome cover with the words Track AI Visibility on a Budget over a faint grid of nodes, a small bar chart and a search answer card.

If you lead marketing at a small team, you've probably been asked whether your brand shows up when people ask ChatGPT or Perplexity about your category, and you may not have a good answer yet. You also may not have a big budget to find one. This guide walks you through a setup for AI visibility under 300 a month, built around a small set of questions, the engines that matter to your buyers, a plan tier that fits, and a review routine one person can keep up with. By the end you'll have a working tracker and a simple habit around it.

You'll need a rough idea of your product category, a few hours to set things up, and a spreadsheet. That's about it.

The short version of the answer is this. Pick a small, fixed set of buyer questions, track them on the engines your audience uses, and look at the same results on a regular schedule. If you're still deciding on tools, the best AI visibility tools roundup gives the wider field. Cheap AI visibility tracking works when you keep the scope tight, because a tracker you can actually read each week is worth more than a big one you ignore.

1. Set your spend ceiling and pick a plan tier

Start with the money, because budget AI monitoring lives or dies on it and it decides everything after it. Write down the most you can spend each month on tracking software, and then decide whether one engine is enough for the question you're trying to answer or whether you need two.

Here is how DeepSmith's plans line up against a $300 ceiling when you pay month to month:

PlanMonthly priceTracked promptsAI engines
Pro$9950ChatGPT
Grow$199100ChatGPT, Perplexity
Scale$399200ChatGPT, Perplexity, Gemini

If ChatGPT alone is enough for you right now, Pro at $99 a month is the simple choice. If you also want Perplexity, Grow at $199 a month is the plan that fits, and it leaves $101 under your ceiling. Scale adds Gemini, but at $399 on monthly billing it sits above $300.

Scale does list an annual rate of $299 a month, and that can look like it fits. It's billed yearly, though, so you'd pay $3,588 up front, and that's not a $299 monthly payment. The annual effective rates for the other plans are $80 for Pro and $160 for Grow, also billed yearly. If your limit is monthly cash going out, stay on monthly billing and pick Pro or Grow. DeepSmith has no long-term contracts and offers a 7-day free trial, so you can try the setup in this guide before you commit to a tier.

You're done with this step when the tier, the billing cadence, the engines it covers and the price are written in one place, and whoever approves the spend has seen them.

Common mistake: treating the monthly equivalent of an annual plan as a monthly charge. Write down what you'd actually pay and when.

If you want to compare other budget-friendly options before you pick, a roundup of affordable AEO tools is a good place to look.

2. Choose a small, stable prompt panel

A tracked prompt is just a question you want to watch AI engines answer. Your plan may allow 50 or 100 of them, but you don't have to fill that allowance. A good place to start is 20 questions, split into four groups of about five each:

  1. Problem discovery, where the buyer is describing a pain and hasn't named a solution yet.
  2. Category and comparison questions, where they're weighing approaches or tools.
  3. Buying criteria, where they're asking what to look for.
  4. Use and implementation, where they want to know how to get something working.

The 20-prompt number is my suggestion for a starting point. It's not a research-backed ideal, and it isn't a limit of any tool. If your team can only review ten questions, use ten.

To find the wording, ask sales, support, product marketing and your content team what buyers actually say. Then use templates like these and swap in your own category terms:

  • "How do I solve [problem] for [audience]?"
  • "What should a [role] look for when choosing a [category]?"
  • "Which tools help [audience] with [specific job]?"
  • "What are the trade-offs between [approach A] and [approach B]?"
  • "What questions should I ask before buying [category]?"
  • "What are the common mistakes when adopting [category]?"
  • "How can [role] measure whether [approach] is working?"
  • "What alternatives to [known solution] should I consider?"

Semrush has a longer write-up of prompt tracking methodology if you want more depth. These are templates, not proof that buyers ask those exact questions, so drop any that don't match your audience. Remove duplicates and tighten vague ones. If you've already got a keyword list, remember that AI prompts differ from keywords, and the questions people type into an assistant tend to be longer and more conversational.

DeepSmith's Discover Prompts can suggest a starter set from your product and persona context, which saves you some blank-page time. Read through the suggestions and keep the ones that fit rather than accepting the whole list.

You're done with this step when each question has an owner, an intent label (which of the four groups it belongs to), and a short note on why you care about the answer. Save a dated copy of the original wording.

Common mistake: rewording a question every time the answer looks bad. A reworded question is a new test, so if you change one, keep both versions and don't compare them as if nothing changed. If you want a deeper walkthrough of this, a repeatable prompt set covers it in more detail.

If you're not sure how many prompts and engines your situation calls for, estimating prompt and engine needs goes into that properly. This guide sticks to one concrete low-cost setup.

3. Pick only the engines your plan covers

It's tempting to want every engine, but a budget AI monitoring setup works better when you choose engines because your buyers use them. On Pro you track ChatGPT. On Grow you track ChatGPT and Perplexity. Scale adds Gemini, and that puts you over the monthly ceiling, as we saw in step one.

Ask yourself where your audience actually goes to ask questions. If you don't know, a quick poll of a few customers and your sales team will usually tell you. Two engines done well is a better use of a small budget than six engines you never look at.

Treat each engine as its own column in your reporting. ChatGPT and Perplexity don't behave the same way, and one engine's result doesn't tell you what all AI search is doing.

You're done with this step when every prompt is assigned to an engine your plan includes, and your report shows which engine produced each result.

Common mistake: comparing one engine's mentions to another engine's citations as though they were the same thing. Keep each engine separate so you can see where you show up and where you don't.

4. Set up your brand and schedule the collection

Now you can put the setup into the tool. DeepSmith's AI Visibility lets you define what counts as your brand, list the competitors you want to compare against, add your stable prompt set, and choose how often collection runs. It keeps the full answer for each run, and it tracks mentions and citations separately.

Take a little care with two settings. First, define your brand name variants, such as the short name people use and the full company name, so that a mention isn't missed because of spelling. Second, keep your competitor list to the companies that really compete with you for the same buyer. Mixing in unrelated companies makes the results harder to read. If you're stuck on who to include, choosing which competitors to track walks through it.

The product page describes daily runs, and for a 20-prompt panel on two engines that works out to 1,200 scheduled observations over 30 days (20 prompts, 2 engines, 30 days). That's simple arithmetic, and it assumes every scheduled check runs. It's not a count the product reports. Look at your own account to confirm the schedule is what you expect, rather than assuming the setting saved.

You're done with this step when every question is being collected on the engine you chose, your brand and competitors are defined, and answers are showing up in the history.

Common mistake: skipping brand name variants, or loading a competitor list full of companies your buyers never consider. Either one makes your numbers look different than reality.

5. Record a baseline and keep mentions and citations apart

Before you change anything on your site, save the first full set of answers. This is your baseline, and every later comparison depends on it.

In a simple worksheet, keep one row per result with these columns: the date or reporting window, the prompt ID and exact wording, the intent group, the engine, whether your brand was mentioned, whether one of your pages was cited, the cited URL if there is one, which competitors showed up, and a short note on the answer. If a check fails or returns something unusable, mark it separately. Don't count it as a "no."

A mention means the answer names your brand. A citation means the answer links to one of your pages as a source. An answer can mention you without citing you, cite you without much of a mention, do both, or do neither, so track the two on their own. If the difference is new to you, AI citations versus brand mentions explains why both matter.

For your own worksheet, a clear way to calculate the rates is to use valid answers as the denominator:

  • Mention rate is the valid answers that mention your brand, divided by all valid answers in the panel and time window.
  • Citation rate is the valid answers that cite one of your pages, divided by all valid answers in the same panel and window.

Here's an example with made-up numbers. If one round gives you 10 valid answers, and 3 mention your brand while 1 cites a brand page, your worksheet rates are 30% and 10%. That's just arithmetic to show the method, not a benchmark or a result from any product.

These are spreadsheet definitions for keeping your own reporting consistent. They don't describe how DeepSmith calculates its numbers internally. For share of voice and period-over-period trends, use the definitions and views the tool displays instead of building your own formula. Its Pages view is also useful here, since it shows which of your URLs get cited, each page's share of your citations, and the prompts behind them. For a wider list of what to measure later, AI visibility metrics and KPIs is worth a look.

You're done with this step when you have a saved baseline and your report clearly separates a brand mention from a link to a brand page.

Pro tip: A score from a tracked prompt set tells you how you did on the prompts you chose. It doesn't describe every AI conversation in your market, so call it your panel visibility and not your total market share. That small change in wording keeps your reports honest when someone senior reads them.

6. Keep the review routine light enough to stick

A routine you skip is worse than a small one you do. The schedule below is a practical suggestion and not a proven requirement, so adjust it to your week.

Weekly, a short scan. Look at changes by prompt and by engine, open a few full answers, and note any new competitor citations. Flag any answer that gets something about your company wrong. A simple weekly note can have five lines:

  1. Which prompts or engines changed in a way worth noting.
  2. New or lost citations to your pages, and which page, if you can see it.
  3. Competitor pages that appeared for your tracked questions.
  4. Any answer that misstates your company.
  5. One next action, who owns it, and the date you'll check again.

Monthly, a longer look. Compare a full 30-day window with the 30 days before it. DeepSmith's visibility reporting describes 30-day metrics and period-over-period change, so use the same window when you compare. Check the list of pages that get cited, and pick one or two findings that are worth acting on. Don't try to act on everything.

Day-to-day swings are normal, and they shouldn't send you chasing every dip. Alert thresholds can help you decide what counts as a meaningful change.

You're done with this step when one person can finish the review without rebuilding the panel or testing every prompt by hand, and your monthly note uses the same prompts, engines and date windows every time.

Common mistake: reporting only one overall percentage. A change might be coming from a single question, a single engine or a single page, and an average hides that. Keep the breakdown so the team can see where it happened.

7. Add Google's reports and your website visits, and keep them separate

Your tracker covers the answers AI engines give. Two other sources round out the picture, and both are free.

For Google's own generative features, Search Console now has generative AI performance reporting. Google announced it in June 2026 and said it had rolled out worldwide by August 31, 2026, as described in its Search Console announcement. It covers impressions and pages, with breakdowns that include country and date. It's a view of Google Search, so it doesn't report on every third-party chatbot answer.

For visits to your site, use GA4. Its default channel group puts traffic from sources like ChatGPT, Gemini, DeepSeek, Copilot and Grok into an "AI Assistant" channel when the medium matches the rule, as laid out in the default channel group documentation. Google AI Overviews and AI Mode don't go into that channel, and they're counted under Organic Search instead. Use the landing page report to see where those visitors arrived. If you want to report on conversions, mark the business action you care about as a key event. Standard reports can take up to 24 hours to show a newly marked key event, and Realtime updates a few minutes after someone triggers it. If you link Search Console data to Analytics, it can take 48 hours after collection to show up.

Put these in separate sections of your report: AI answer visibility from your tracker, Google generative-feature impressions from Search Console, and referral visits or conversions from GA4. They're related, but they measure different things, so don't add them together and don't say one caused another without evidence.

You're done with this step when your report has those three sections and each one names its source and date range.

Common mistake: treating a visit from an AI assistant as a count of every time your brand appeared in an answer. Many answers never lead to a click, and Search Console's Google AI reporting only covers Google's own surfaces.

What to do next

Once the setup is running, give it a full month before you judge it. The first pass is the baseline, and the second month is when comparisons start to mean something. At that point you can decide whether to add prompts, add an engine or move up a tier.

If you want to see how this fits into your wider budget, setting an AEO budget covers how to split spend across tracking, content and fixes. And if you're still weighing whether a tool beats a spreadsheet at your size, a tool versus a DIY spreadsheet lays out when manual tracking stops scaling.

When you're ready to try it, you can start a free DeepSmith trial, load your 20 prompts, and have a baseline by the end of the week.

Frequently asked questions

Can I track AI visibility for under $300 a month?

Yes. Based on listed monthly prices, DeepSmith Pro is $99 a month for ChatGPT, and Grow is $199 a month for ChatGPT and Perplexity. Pick the engine coverage that matches the question you're trying to answer.

Which DeepSmith plan includes Gemini, and does it fit under $300?

Scale includes Gemini, but it's $399 on monthly billing. Its annual effective rate is $299 a month, and that requires paying for the year up front, so it isn't a $299 monthly charge.

How many prompts should I start with?

Twenty fixed prompts, grouped by buyer intent, is a practical starting point. It's a suggestion and not a research-backed ideal. Use fewer if your team can't review that many, and don't fill the plan allowance just because it's there.

Does a brand mention mean the AI cited my website?

No. A mention means the answer names your brand, and a citation means it links to one of your pages as a source. An answer can do one without the other, so track them separately.