Want more qualified Shopify traffic from AI search?
Run a free Naridon scan to see which prompts, products, and AI engines can send more ready-to-buy visitors.
Most merchants check their AI visibility exactly once. They open ChatGPT, type a question, see their product named, and feel great, or do not see it and panic. Both reactions are built on the same mistake: a single run is one sample of a moving target. Ask again with different phrasing, or next week, or from a different account, and the answer can change completely. A screenshot is not measurement.
Knowing whether AI recommends your products is a tracking problem, not a spot-check. This post lays out how to measure it systematically, and how to act on what you find so the competitors currently getting named stop showing up in your place.
Why one check tells you nothing
AI answers are not static rankings. They vary by exact phrasing, by whatever context the user has, and over time as engines re-crawl and re-reason. So the product named for “best waterproof hiking boots” might differ from the one named for “durable waterproof boots for wet trails,” and both might change next month. Judging your visibility from a single query is like judging your search rankings from one search on one device: technically a data point, practically misleading.
The method: prompt set, cadence, share of voice
Build a fixed prompt set
Start with the questions your shoppers actually ask, the buyer-intent, natural-language prompts that lead to a purchase in your category. Include comparisons, use cases, budgets, and constraints. This set becomes your benchmark, so keep it stable enough to trend and expand it deliberately as you learn.
Run it on a schedule
Run the whole set across the engines on a regular cadence, weekly if you can, monthly at least. Consistency is the point: the same prompts at the same interval let you separate real movement from noise and tie changes to what you shipped. One run is a sample; a series is a trend.
Score share of voice
For each prompt, record whether your products are named, whether a competitor is named, and whether the mention links to you. Roll that up into share of voice, the percentage of prompts where you are named, ideally weighted by prompt importance. Because it measures the answer itself, share of voice is immune to the referral misattribution that plagues click-based metrics, covered in why ChatGPT referral traffic is missing from your analytics. It is the cleanest number you have.
Track every engine, because they disagree
The engines answer differently, and a competitor may own one while you own another. You might be strong in Perplexity and invisible in Gemini, and you would never know from checking only ChatGPT. Running the same set across ChatGPT, Perplexity, Google AI Overviews, Gemini, and Copilot on the same schedule gives you a per-engine breakdown so you can prioritize the surface where you are weakest. Each engine has its own levers, detailed in guides like getting recommended by Gemini and getting recommended by Copilot.
What to do with a prompt you are losing
Every prompt where a competitor is named and you are not is a work item, not a verdict. The loop:
| Step | Action |
|---|---|
| 1. Diagnose | Read the answer, note which sources were cited and what made them quotable |
| 2. Fix the page | Repair schema, state specific facts, add an FAQ that answers that exact question |
| 3. Check access | Confirm the engine's crawler can reach the page and nothing blocks it |
| 4. Corroborate | Earn a third-party mention for the use case if the cited sources have one |
| 5. Re-measure | Re-run the prompt next cycle to confirm you moved into the answer |
That detect, fix, re-measure loop is the entire discipline. Do it across your prompt set and share of voice climbs.
Manual or tool?
By hand is fine to build intuition and fine for a handful of prompts. It breaks down the moment you want dozens of prompts across five engines every week, because you cannot do that consistently without drift, and inconsistency destroys the trend you are trying to build. A tool gives you the same prompts, the same cadence, and a recorded history, turning anecdote into measurement.
How Naridon does it
Install Naridon free from the Shopify App Store, free forever at $0 with 150 credits a month. It runs your buyer prompts across five engines on a schedule, reports share of voice, citations, and sentiment per engine, and shows you exactly which prompts a competitor is winning. Then it closes the loop: Autopilot writes the schema, copy, and FAQ fixes, verifies each with an LLM before publishing, applies them to your catalog, and re-measures whether you moved into the answer. Measurement and fixes in one place. Paid plans from $49/mo with a 7-day trial. See the pricing page.
Knowing whether AI recommends your products is not a screenshot, it is a tracked share-of-voice number on a fixed prompt set across every engine. Build the set, run it on a cadence, treat each lost prompt as a fix, and re-measure. That is how you turn a lucky mention into a visibility trend you actually control.
Frequently asked
- How do I know if AI is recommending my Shopify products?
- Do not rely on asking once and eyeballing it, because answers vary by phrasing, by user, and over time. Instead, build a fixed set of the prompts your shoppers actually ask, run them across the major engines on a schedule, and record for each whether your products are named, whether a competitor is named, and whether the mention links to you. That gives you a share-of-voice number you can trend instead of a lucky screenshot. A monitoring tool automates the runs so you measure consistently rather than sampling on a good day.
- What is share of voice in AI search?
- Share of voice is the percentage of your tracked buyer prompts where an AI engine names your products, ideally weighted by how important each prompt is. It is the cleanest single measure of AI visibility because it looks at the answer itself rather than downstream clicks, so it is not distorted by referral misattribution. Rising share of voice is the leading indicator that predicts AI referral, branded search, and direct traffic. Track it per engine and in aggregate to see where you win and where a rival owns the answer.
- How often should I check my AI product visibility?
- On a regular schedule rather than ad hoc, because AI answers drift as engines re-crawl, competitors change, and phrasing varies. A weekly or at least monthly cadence on a fixed prompt set lets you separate real movement from noise and tie changes to the fixes you ship. Checking once and declaring victory or defeat is the mistake, one run is a single sample of a moving target. Consistent cadence on the same prompts is what turns spot checks into a trend you can manage.
- Why does AI recommend my competitor instead of me?
- Usually because their pages are easier for the engine to quote for that specific question: clearer factual copy, complete Product and Review schema, more parseable reviews, third-party mentions for the exact use case, and open crawler access. It is rarely paid placement. The productive response is to find the prompts where a competitor is named, study why their source is more quotable, and close the gap on your own pages. Tracking which prompts name whom over time is the part you cannot eyeball, which is where a monitoring tool earns its keep.
- Can I track AI recommendations across ChatGPT, Perplexity, and Gemini at once?
- Yes, and you should, because the engines answer differently and a competitor may own one while you own another. A monitoring tool runs your prompt set across multiple engines on the same schedule and reports share of voice per engine, so you can see, for example, that you are strong in Perplexity but invisible in Gemini and prioritize accordingly. Naridon tracks five engines, ChatGPT, Perplexity, Google AI Overviews, Gemini, and Copilot, in one place so the comparison is apples to apples.
- What should I do when I find a prompt where I am not recommended?
- Treat it as a work item. Read the answer, note which sources were cited and why they were quotable, then fix the corresponding gap on your pages: add or repair schema, make the product copy state specific facts, add an FAQ that answers that exact question, ensure crawler access, and earn a corroborating mention if the cited sources have one. Then re-run the prompt on your next cycle to confirm the fix moved you into the answer. That detect, fix, re-measure loop is the entire game.
- Is manual prompt checking good enough, or do I need a tool?
- Manual checking is fine to start and to build intuition, but it does not scale and it is not consistent: you cannot rerun dozens of prompts across five engines every week by hand without drifting. A tool gives you the same prompts, the same cadence, and a recorded history so you can trend share of voice and attribute movement to your work. If you have a handful of prompts and lots of patience, do it by hand; beyond that, automate it so the measurement is reliable rather than anecdotal.
- How does tracking AI recommendations connect to sales?
- Share of voice is a leading indicator: when the engines start naming your products for buyer-intent prompts, AI referral, branded search, and direct traffic tend to follow a few weeks later, and sales with them. So tracking recommendations is not vanity, it is the earliest signal that your GEO work will pay off, well before it shows in revenue reports. Watch share of voice rise first, then confirm the downstream lift in your Shopify and GA4 analytics.
Key concepts
Plain-language definitions of the terms in this guide.
Ready to rank for these conversations?
Join early adopters who are already capturing AI search traffic.