If you lead marketing and someone has asked what your AI search plan is, one of the first questions you hit is which AI engines to track. There are a lot of them now, and most AI visibility tools cover only some, so you can't track everything on day one. This guide walks you through a simple way to rank the engines for your own business, using your audience, your referral data and the way your buyers actually ask questions. By the end you'll have a short list of two or three engines to start with, and a clear note on what would make you change it.
Before you start, it helps to have a few things on hand: access to your web analytics, a rough idea of who your buyers are, and a spreadsheet. If you're also still deciding on software, you can read about AI visibility tools separately, because the ranking below works with any of them, or with no tool at all.
One thing to say up front. Nobody has published a ranking that tells every brand to track the same three engines, so be wary of any article that claims to. What the research supports for choosing which AI engines to track is a starting guess (ChatGPT is a reasonable broad-reach first candidate) and then a method for using your own evidence to pick the next one or two. That's what the steps below are for.
Why the question matters at all
Buyers are using AI alongside Google, and in some cases ahead of it. In a March 2026 G2 survey of 1,076 B2B software buyers and decision-makers, 51% said they now start research with an AI chatbot more often than Google, up from 29% in G2's 2025 report. Also in that survey, 71% said they relied on AI chatbots at some point in their research, and 61% said they use AI search alongside Google. So tracking AI search makes sense as an addition to what you already do, not a replacement.
The survey doesn't say which chatbots those buyers used, though, so it can't tell you which engine to start with. That part is yours to work out, and it's why a method is more useful than a list.
Step 1: Set the business outcome you want to improve
Pick one main outcome for this exercise. It could be showing up in the answers buyers see, getting qualified visits, getting leads or purchases, or getting one area of your content cited. Then write down who you're trying to reach, which country or language, which product, and which stage of the buying process matters most.
You're done with this step when your team can say what a useful result from an engine would look like, for whom, and why. For example, "we want our product named when an ops manager in the US asks for tools in our category" is specific enough to work with.
Common mistake: choosing an engine because it's popular without checking whether that audience is yours, or without knowing whether you care about mentions, citations, referral visits or conversions. Those are four different goals, and they can point to different engines.
Step 2: Map how your buyers ask for help
Engines matter only in relation to the questions people put to them, so list those questions before you look at any engine. Start with what sales and support hear most, what your customers say in interviews, what people search for on your own site, and the topics your existing content covers. If you want a longer walk-through, here's how to find the prompts to track for a brand.
Aim for a compact set that covers awareness (learning about the category), consideration (comparing approaches), and decision (evaluating vendors). Keep each question tied to your category and use case, since a generic list of AI questions won't tell you anything about your buyers.
You're done when you have a representative set of buyer questions across all three stages. A few dozen is plenty for this exercise.
Common mistake: only looking at top-of-funnel questions. The B2B buyer surveys describe AI use across many parts of research and purchase, so decision-stage questions belong in the set too.
Step 3: Find out which surfaces your audience actually uses
Now ask which AI products those buyers reach for. The best sources are the ones closest to you: short interviews or a one-question survey for customers and prospects, notes from sales calls, and your analytics. Ask which AI products they use for research, which for comparing vendors, and which for work tasks. Where possible, separate a product's web-search mode from general chatting, because an answer that cites the web behaves differently from one that doesn't.
Keep Google's search features apart from standalone chat apps. Google's own Search Central guidance says AI Overviews and AI Mode can use different models and techniques, and the Gemini app is a separate product from both. If Google Search is central to how your buyers find you, it's worth reading how AI Overviews and AI Mode differ for tracking before you decide how many Google surfaces to count.
Public data can give you a rough sense of reach, as long as you treat it as background and not as an answer. A Pew Research Center survey of 5,123 US adults in early 2025 found that 34% had ever used ChatGPT, and 58% of adults under 30 had. Statcounter's worldwide AI chatbot chart for October 2026 lists ChatGPT at 51.58%, Gemini at 28.86%, Perplexity at 15.04%, Claude at 2.3%, Microsoft Copilot at 2.18% and DeepSeek at 0.05%. The chart doesn't spell out what those percentages are a share of, so don't read them as shares of users, queries, referrals or buyers. If you want a wider look at how the engines compare on audience, there's a piece on AI search engines ranked by users and momentum.
You're done when each engine you're considering has a reason from your own buyers, or is clearly marked as a guess because you don't have direct evidence yet.
Common mistake: treating a broad market-share chart as a measure of your customers. A chart can say ChatGPT is the biggest overall and still be wrong about where your particular buyers go. The same goes for treating every assistant from one company as one interchangeable engine. Gemini, AI Overviews and AI Mode are three different things to track.
Step 4: Measure referrals and outcomes by source
Next, look at what your own analytics already show. In your analytics and CRM, pull the visits that came from AI products, along with the landing pages they reached, how long they stayed, and whether they turned into leads or sales. If you use Google Analytics 4, the default AI Assistant channel groups sources such as ChatGPT, Gemini, DeepSeek, Copilot and Grok, and Google's description excludes AI Overviews and AI Mode from it. That means it's a useful view and not a full record of AI exposure. There's a longer guide on how to measure AI search traffic in GA4 if you need the setup.
Compare like with like. Use the same time period, the same country, the same definition of a conversion, and the same denominator for every source. Keep the session's source separate from the source credited for a key event, because GA4 treats those as different dimensions and mixing them will muddy the numbers.
Some outside data suggests the visits can be valuable. Adobe reported that in September 2026, AI traffic to US retail sites grew 71% year over year and converted to purchases 90% better than non-AI traffic, and that its July report put the gap at 60%. Those are different months, and both are retail-wide numbers, so they say nothing about your brand or about any single engine. Treat them as a reason to look at your own data, not as a benchmark to expect.
You're done when you have a small table of visits and outcomes by source, including "no measurable referrals" where that's what you found, and a note on where attribution has gaps.
Common mistake: reading a low referral count as proof that an engine never influences a buyer. Referral data only captures people who clicked a link. An engine can name your brand to a buyer who later searches for you directly, and that never shows up as a referral.
Step 5: Score each candidate with one clear rubric
Now put the evidence in one place. Give each candidate engine a rating from 0 to 5 on four criteria, and weight them like this. The weights are our suggestion for a practical starting point and not the result of any published study, so adjust them to fit your goals.
| Criterion | Suggested weight | Evidence to use |
|---|---|---|
| Audience fit | 35% | Which engines your buyers say they use, your market and language, and how they actually do research |
| Referral and outcome evidence | 25% | AI-attributed sessions, qualified visits, leads or purchases, read with the attribution caveats from step 4 |
| Buying-journey relevance | 25% | Whether the engine looks relevant to the real questions and decisions in your prompt set |
| Coverage and operational fit | 15% | Whether you can track it on your plan or with another reporting source, and whether it adds coverage the others don't |
To get a score, multiply each rating by its weight and add them up. Keep the raw evidence right next to each number so a reviewer can see why an engine scored the way it did. Where you have no evidence for a criterion, write "unknown" and leave it out of the average for that engine, instead of quietly giving it a high score. A score is a decision aid and not a precise measurement.
A few notes on what to expect for specific engines, taken from what's publicly known:
- ChatGPT: a sensible first candidate for a broad pilot, since both the Pew survey and the Statcounter chart point to wide reach. ChatGPT can search the web and show citations, but OpenAI itself warns that results and citations can be incomplete, outdated or incorrect, so check any metric against your own buyer questions.
- Gemini: a strong early comparison where Google-related audiences matter. Keep it separate from AI Overviews and AI Mode.
- Perplexity: worth including early when your buyers do a lot of web research and like source-linked answers, or when your own referral data supports it. Perplexity describes itself as a web-search product whose answers include citations and links.
- Claude: consider it when your buyer research or analytics says it matters. Anthropic says Claude searches the web when applicable and gives citations when it does, though that doesn't mean every reply uses a search.
- Google AI Overviews and AI Mode: count them separately when Google Search is central to your audience. Google offers a generative AI performance report in Search Console (rolled out worldwide on August 31, 2026), which can supply some of your evidence without any extra tool.
- Microsoft Copilot: prioritize it if your audience, referrals or buying process point to it. Copilot Search lists its sources and links.
- Grok, Meta AI and DeepSeek: no reliable buyer-prevalence or conversion benchmark turned up for these, so score them with the same first-party evidence as the rest, without assuming a ranking.
You're done when every engine on your list has a score and a short note on what is observed versus assumed.
Common mistake: giving a made-up score the look of scientific certainty. If most of your ratings are guesses, say so in the sheet. That honesty is what makes the ranking useful when someone later asks why you picked what you picked.
Step 6: Choose the first two or three, and write down what would change your mind
Take the highest-scoring engines, then check that the short list covers different buyer behaviors and not three near-copies of one another. If you have no brand-specific data yet, use ChatGPT as your starting hypothesis and compare Gemini and Perplexity for the next slot, based on buyer usage, referral evidence and the prompts that matter to you. That is the heart of the ChatGPT vs Gemini tracking priority question: the answer depends on whether your buyers lean on a general assistant, a Google-connected one, or a source-linked search tool, and your evidence from steps 3 and 4 settles it better than any chart. Add a Google Search AI surface as its own item when it fits your audience and you have a way to measure it. Treat Claude, Copilot, Grok, Meta AI and DeepSeek as candidates when your evidence supports them, and neither automatic inclusions nor automatic exclusions.
Two or three engines is a good first number because it leaves you enough attention to read the results. That's a practical suggestion for a first pass and not a published threshold, so widen it later if your evidence and budget allow.
Then write a short decision note. It should name the two or three engines, give the evidence for each, list the ones you left out, and say what would trigger a re-ranking, such as a jump in referrals from an engine you skipped, or a customer interview that keeps naming one. If you later want a deeper setup for a chosen engine, the guides on tracking Gemini and tracking Perplexity go through each in detail, and they're outside the scope of this piece.
You're done when the note exists and someone outside your team could follow the reasoning.
Pro tip: your plan limits how many engines you can follow, so check coverage against your short list before you commit. DeepSmith's Pro plan tracks ChatGPT, Grow adds Perplexity, Scale adds Gemini, and Enterprise covers all ten engines (ChatGPT, Gemini, Perplexity, Claude, Google AI Overviews, Google AI Mode, Grok, Meta AI, Microsoft Copilot and DeepSeek). Matching the list to the plan is part of the AI engine coverage decision, and it's easier to do now than after you've paid.
Common mistake: declaring one engine the universal winner from a cross-industry chart or one company's retail conversion result.
Step 7: Run a small pilot and revisit the ranking
Now test the ranking. Track the same representative buyer questions across your chosen engines, record mentions and citations separately, and compare how they change over a steady period. A mention means the engine names your brand, and a citation means it links to one of your pages as a source, so a result can be one without the other. If you want a plain walk-through of running one question set on several engines, here's how to track the same prompts across engines. The size of the set matters too, and there's a guide on how many prompts to track if you're unsure.
This is a step where a tool saves time. DeepSmith checks the questions you define on a schedule, captures the answers, and reports mention rate, citation rate, share of voice and visibility trend, with a per-platform breakdown so you can compare engines side by side. It handles the answer-level part of the work. Your own analytics and CRM still tell you what happened after a click, and DeepSmith doesn't replace them.
Pair the answer-level results with the referral and outcome data from step 4 whenever you can. Pick a review date in advance, such as 30 or 90 days. Those windows are our suggested planning cadence and not a proven threshold, so choose what gives you enough observations to separate a real pattern from noise.
When you review, sort the engines into three groups: the ones that earned continued coverage, the ones that are still uncertain, and the ones to drop or leave out. If an engine shows you in one place and not another, that's worth a closer look, and a piece on per-engine visibility gaps explains how to read it.
You're done when you can say which engine earned its place, which is still open, and what evidence will drive the next budget or coverage decision.
Common mistake: treating one answer or one week as a stable ranking. AI answers vary from run to run, so a single result tells you very little, and mixing up mentions with citations to your pages will make the picture look better or worse than it is.
What to do next
Open a spreadsheet today and do steps 1 and 2 in an hour: write the outcome you want and list twenty or so real buyer questions. That alone gives you something to test, and the rest of the steps are mostly filling in evidence you already have in your analytics. If you'd like to see mention and citation rates across several engines without building the tracking yourself, you can start a DeepSmith free trial and run your first question set.
Related guides
- Getmint Alternatives: 8 Best Answer-Engine Visibility Tools for 2026: Compares eight answer-engine visibility tools as Getmint replacements, each with a best-fit use case, weighing engine coverage, prompt limits, pricing, and reporting cadence, aimed at buyers whose current.
- Getvisoryn Alternatives: 8 Best AI Mention Monitoring Tools for 2026: Reviews eight AI mention monitoring tools as Getvisoryn replacements (DeepSmith, Scrunch, Profound, OtterlyAI, Peec AI, Rankscale, Goodie, ZipTie), comparing pricing, engine coverage, and what a buyer can actually.
- Goodie AI Alternatives: 8 Best GEO and AI Visibility Platforms for 2026: A buyer roundup of eight generative-engine-optimization platforms you could switch to from Goodie AI (DeepSmith, Profound, Peec AI, OtterlyAI, LLMrefs, Semrush, Foglift, AthenaHQ), comparing published pricing, engine coverage.
- LLMpulse Alternatives: 8 Best AI Search Visibility Platforms for 2026: Explains what LLM Pulse covers and its published euro pricing, then ranks eight replacements (DeepSmith, Profound, AthenaHQ, Scrunch, LLMrefs, Rankscale, Goodie, Otterly) by engine coverage, prompt allowances, cost.
- Maxaeo Alternatives: 8 Best Multi-Engine AEO Trackers for 2026: For buyers who want broad AI engine coverage, this ranks eight Maxaeo substitutes (DeepSmith, Rankscale, Scrunch, OtterlyAI, Peec AI, Goodie, Profound, ZipTie.dev) on engine breadth per self-serve plan.


