Guide
AI search monitoring for ecommerce: what it is and how to set it up
Updated
AI search monitoring is checking, on a schedule, what AI answer engines say when shoppers ask them what to buy: whether they name your store, whether they link to it, which competitor they name instead, and whether the price, stock and policies they quote are right. It is not rank tracking. There is no results page to rank on; an engine names a few brands or none, the list changes from one run to the next, and the honest measurement is how often you appear across repeated samples, not where.
What should an online store monitor?
Five things, each per prompt and per engine, each recorded with a date and a sample count. Definitions for the terms are on the glossary.
- 01
Mentions
Whether the answer names your brand at all, per prompt and per engine. This is the recommendation signal. Track it as a frequency: named in 4 of 6 samples this scan, 2 of 6 last scan.
- 02
Citations
Whether the answer lists one of your own pages as a source. This is the stronger signal and the one that sends traffic. An answer can cite your page without naming you (a ghost citation) or name you without citing you; keep the two apart.
- 03
Who is named instead
On the prompts where you are absent, which brand the engine recommends. Share of Voice against named competitors turns a vague sense of losing into a list of prompts and a name.
- 04
Accuracy of price, stock and policies
Engines quote prices, availability, shipping and return terms, and they are sometimes wrong. A confident wrong answer about your store costs orders you never see. Check what is said against what is true, per engine, because one engine being right does not mean the others are.
- 05
AI-referred traffic and orders
Visits arriving from chatgpt.com, perplexity.ai, gemini.google.com, claude.ai and copilot.microsoft.com, and the orders those visitors go on to place. These referrals often land on a product page with no campaign tag, so they need first-touch attribution rather than last-click.
Why don't Shopify's reports show it?
Shopify's acquisition reports group sessions by referring channel, platform and medium, with examples like Instagram, Google and Facebook and mediums like social, search and email. There is no AI category, so a visit from chatgpt.com is a referral like any other and a visit from an AI answer that opened a product page directly has no campaign tag at all. Merchants who want the number filter referrer domains by hand, and even then the report credits the order to whichever visit came last.
The other half of monitoring, what the engines actually say, is not in any store report, because it never touches the store: the shopper asked an engine, the engine answered, and the store only finds out if the shopper clicks. That is why the answers themselves have to be collected, on a schedule, through the engines' APIs.
How do you set it up in a week?
A working setup needs a prompt list, a competitor list, a dated baseline and a way to see referrals. Seven steps, one a day, and the last one is the one most stores skip.
- 01
Write 10 to 25 buyer-intent promptsDay 1
Questions shoppers ask when choosing, not your brand name: "best soy candles for gifts", "linen vs cotton sheets", "is a weighted blanket worth it". Mix best-of, comparison and worth-it questions across your top categories. Buyer-intent prompt, defined.
- 02
List three to five competitors by nameDay 1
The stores a shopper would buy from instead. Include the aliases and spellings each brand is known by; a competitor you have not listed is not in the denominator.
- 03
Take a baseline through the engines' official APIsDay 2
Run every prompt on every engine you care about, more than once each, and record the answer text, the cited URLs, the model identifier and the date. Do not script the consumer chat apps; their terms forbid it and their answers are not reproducible. A free scan gives a dated baseline without an account. Free AI shopping scan.
- 04
Check crawler access and index eligibilityDay 3
Confirm robots.txt allows OAI-SearchBot (ChatGPT search), Claude-SearchBot and PerplexityBot, whatever you decide about training crawlers such as GPTBot. Confirm product and collection pages are indexed and snippet-eligible: noindex, nosnippet and max-snippet:0 remove a page from Google's AI features as well as from classic results. OAI-SearchBot vs GPTBot.
- 05
Verify the facts engines quote about youDay 4
Ask each engine your price, shipping, returns and whether the store is open, and compare with the truth. Write down what was wrong, where, and on which date; that record is what you will need if it stays wrong. Free AI accuracy scan.
- 06
Tag AI referrers and use first-touch attributionDay 5
Shopify's acquisition reports have no AI channel, so filter the referrer domains yourself (chatgpt.com, perplexity.ai, gemini.google.com, claude.ai, copilot.microsoft.com) and credit an order to the AI visit that came first, not the last click. How RevenueBird attributes orders.
- 07
Set a cadence and alert thresholdsDay 6
Weekly is enough to see movement for most stores; daily if you run promotions engines might misquote. Alert on a lost mention or a competitor taking a prompt you held, not on single-sample changes.
- 08
Write the baseline downDay 7
Date, prompts, engines, models, sample counts, and the frequencies. Engine-side changes can move every brand's numbers at once; without a dated baseline you cannot tell your change from theirs. PositionBird's measurement method.
What does good look like?
A frequency that rises over time, with the sample count next to it. "Named in 3 of 10 samples in August, 6 of 10 in September, on ChatGPT and Claude but not Gemini" is a result a merchant can act on. "Ranked #2 in AI" is not, because the next run will produce a different list. Read the score over several scans, watch the per-engine split (engines retrieve from different indexes, so they move separately), and treat a few points of movement between two scans as noise.
The second mark of good is a shorter list of prompts where a competitor is named and you are not, and a citation rate that climbs toward the mention rate: being cited is the engine sending shoppers to your page rather than describing you from someone else's. The third is zero open accuracy problems, checked per engine. GEO vs SEO vs AEO covers which layer each fix belongs to.
Which tools do this?
Published prices as shown on each vendor's pricing page on the date noted, in USD per month unless stated. "Publishes methodology" refers to the CitedIndex census of 74 AI-visibility tools (2026-08-06), which found 6 with a checkable methodology; PositionBird was not among the tools surveyed and reports its own answer directly.
| Tool | Published price (USD/mo) | Built for | Publishes methodology | Source |
|---|---|---|---|---|
| PositionBird | Free $0 · Starter $19 · Growth $49 · Scale $79 | Shopify stores and any website; mentions, citations, Share of Voice, accuracy checks, AI-referred orders | Yes: engines, models, sampling, formula and limits on /methodology | Pricing |
| Otterly AI | Lite $29 · Standard $189 · Premium $489 · Enterprise from $1,000 | Brands and agencies; prompt-based monitoring across engines | Not among the 6 tools the CitedIndex census (Aug 2026) found publishing a checkable methodology | otterly.ai/pricing, 2026-09-04 |
| Profound | Starter $99 · Growth $399 (billed yearly) · Enterprise custom | Enterprise brands and agencies | Not among the 6 tools the CitedIndex census (Aug 2026) found publishing a checkable methodology | tryprofound.com/pricing, 2026-09-04 |
| Trakkr | Growth $100 · Scale $500 · Enterprise from $790 (billed annually) | Brands and agencies; large prompt and glossary library | Not among the 6 tools the CitedIndex census (Aug 2026) found publishing a checkable methodology | trakkr.ai/pricing, 2026-09-04 |
| Peec AI | Not published (annual plans; sales contact) | Brands and agencies | Not among the 6 tools the CitedIndex census (Aug 2026) found publishing a checkable methodology | peec.ai/pricing, 2026-09-04 |
Otterly AI, Profound, Trakkr and Peec AI are trademarks of their respective owners. PositionBird is not affiliated with or endorsed by any of them. Prices and plan names change; check the linked pages.
Questions stores ask
Is AI search monitoring the same as rank tracking?
No. Rank tracking records the position of a URL on a results page that is the same for everyone who searches. AI search monitoring records whether a generated answer names or cites a brand, and the answer changes from run to run, so the measurement is a frequency across repeated samples (named in 7 of 10 runs) rather than a position. A tool that reports a single rank inside AI answers is reporting one sample of a distribution.
How often should a store check?
Weekly for most stores, with several samples per prompt and engine each time; daily if prices or stock change often enough that a stale answer costs orders. What matters more than the interval is consistency: the same prompts, the same engines and the same sample count each time, so scans are comparable. PositionBird runs monthly on the free plan and up to daily on paid plans.
Can a store see AI referrals in Shopify's reports?
Not as a channel. Shopify's acquisition reports group visits by referring channel, platform and medium (search, social, email, direct) and have no AI category, so visits from chatgpt.com or perplexity.ai have to be filtered by referrer domain by hand, and orders placed on a later visit are credited to that later visit. PositionBird's RevenueBird module records the AI referral at first touch and credits the order to it within seven days.
Does llms.txt help a store get recommended?
There is no evidence that it does. No major engine has committed to reading llms.txt, and in an Ahrefs study of 137,210 domains (June 2026) 97% of published llms.txt files received no requests at all. Shopify generates one for every store automatically, so there is nothing to add. Crawler access, indexable product pages and accurate third-party listings are where measured differences come from.
When an engine already has it wrong
Monitoring finds the problem; these guides cover the fix, engine by engine:
Start with a dated baseline
Free plan available, no card. Or see plans and limits for weekly and daily cadences.
Get it on the App Store