See Gravton in action for your Store! 🚀 Launching Self Serve

See Gravton in action for your Store! 🚀 Launching Self Serve

See Gravton in action for your Store! 🚀 Launching Self Serve

How Much Revenue Are You Losing to AI Invisibility?

Share this article

Share this article

Explore with AI

How Much Revenue Are You Losing to AI Invisibility?

How Much Revenue Are You Losing to AI Invisibility?

AI invisibility, a product missing from an AI answer, is not automatically lost revenue. The loss worth estimating is demand at risk: commercially relevant shopper demand where your products genuinely fit but rarely enter AI consideration, sized as a range of scenarios rather than a single loss figure.

Tarun Babbar

Grounding in AI search
Grounding in AI search

Our five-brand sample shows where that risk tends to sit. The brands appeared in every answer to questions that named them, but in only 20 of 52 answers (38%) to problem-led questions, where shoppers describe a need, such as a skin concern or a fit problem, and leave the AI to choose the brand. That matters because Adobe found that 53% of AI shopping use happens at the top of the funnel, during inspiration and discovery (Adobe).

Key takeaways

  • Count a gap only when three things hold: real demand, genuine product fit and weak AI participation.

  • Size demand at the intent-cluster level, from evidence outside your prompt set.

  • Report a low, base and high range, labeled as exposure, never as revenue lost.

  • Keep "the product was not recommended" apart from "the product was recommended but sold elsewhere".

Which AI visibility gaps count as demand at risk?

Start with product fit. Suppose shoppers ask for a waterproof hiking jacket under $150 and your closest product costs $280. If it does not appear, the AI is right: the assortment fails the price constraint, and calling the absence a loss would overstate the opportunity.

Now take a lightweight waterproof shell under $300 for broad shoulders. You sell a $260 jacket that fits, is in stock and competes with the products being recommended. That absence deserves investigation.

OpenAI says ChatGPT selects products that it judges relevant to the shopper's intent (OpenAI Help Center), so a product that misses a stated constraint should lose. Your data also has to reflect reality, since Google warns that wrong or missing product information can limit a product's eligibility (Google Merchant Center Help).

Our scorecards hold real cases of the second kind. For the home goods brand, its best-selling product sat inside the price ceiling of a question asking for the best product of exactly that type, yet neither engine named it. Fit was there and participation was not, which is exactly the pattern demand at risk is meant to catch.

How we measured this. We tagged every question in our October 2026 scorecards by type: problem-led, price-anchored, occasion or attribute, alternative, where to buy, comparison and worth it. The five brands (two beauty, two jewelry, one home goods) each had 20 questions, answered once by ChatGPT and once by Google AI Mode. They reached us through sales conversations rather than at random, and some question types have only a few answers, so treat the splits as directional.



Related resources

Continue with these related Gravton resources:

How do you show that the demand matters?

A severe gap can still be commercially trivial.

Imagine two needs where your products perform equally badly. The first shows up in onsite search, Search Console, category research and past sales. The second appears mainly because a prompt-generation exercise produced many variations of one idea. The AI result looks the same; the commercial importance does not.

So demand evidence has to come from outside the monitoring set: first-party site search, Search Console (whose Performance report shows how often your site appeared and was clicked in Google Search results, Search Console Help), product and category traffic, past sales, support signals, marketplace research and seasonal patterns. Google's Merchant Center AI performance insights help here because they keep demand and participation apart. "Frequency" is "a representation of how popular specific search types, search terms, search intent, or attributes have been", while a separate measure counts how many of your products appear for them (Google Merchant Center Help). The report covers organic AI traffic on Google only, so use it as one signal among several, not as total market demand.

Why should demand be sized by intent cluster?

Because wording multiplies faster than demand: "stability shoes for wide feet" and "running shoes for overpronation with a wide fit" describe largely the same problem, and valuing each phrasing separately inflates the estimate. Use the shopper need, or intent cluster, as the unit, and give each cluster three fields: product fit (do we have products that genuinely satisfy it?), demand evidence (how important does it look from independent sources?) and AI participation (how often do our eligible products enter consideration today?). Only clusters with all three become candidates for sizing. Our guide to what shoppers ask AI covers how to build the clusters themselves.

How should you rank the candidates?

Rank on all three fields together, plus confidence, never on the visibility gap alone. Confidence is high when demand evidence is first-party, fit is obvious and weak participation repeats across runs and engines; it is low when demand rests on synthetic questions or a handful of answers. Record it before you calculate any dollars, and let it widen or narrow the range.

Shopper need

Product fit

Demand evidence

AI participation

Confidence

Decision

Wide-foot walking

Strong

High

Weak

High

Act now

City travel comfort

Strong

Medium

Weak

Medium

Investigate

Formal event sandal

Strong

Low

Very weak

Medium

Monitor

Sandal under $75

None

High

Weak

High

Exclude: no product fits

Illustrative clusters for a fictional footwear brand.

The formal-event cluster has worse AI performance than city travel but weaker demand, so it ranks lower. The under-$75 cluster has real demand and poor visibility, yet it drops out because the assortment cannot meet it.

How do you size the exposure without calling it lost revenue?

Use scenarios. A practical model multiplies qualified demand (monthly searches or sessions for the need, from your own and category data) by an assumed AI-assisted share of that demand, the participation gain you think you can recover, the share of wins your preferred store captures, the rate at which a recommendation turns into an order, and average order value. Label every input as observed, independently evidenced or assumed. The AI-assisted share, for example, is an assumption; Adobe's July 2026 survey found that 42% of more than 5,000 US consumers had used AI assistants for online shopping (Adobe), which says AI is part of many journeys but not what share of one cluster's demand it touches.

Take an illustrative footwear brand with a $160 waterproof trail shoe in wide sizes, and a cluster around "waterproof trail shoes for wide feet under $180":

Input

Low

Base

High

Relevant monthly demand

40,000

50,000

60,000

Assumed AI-assisted share of demand

15%

20%

25%

Recoverable participation gain

10 points

15 points

20 points

Preferred-store capture

50%

55%

60%

Recommendation-to-order rate

2.5%

3.0%

3.5%

Average order value

$160

$160

$160

Monthly revenue exposure

$1,200

$3,960

$10,080

Illustrative assumptions, not benchmarks. Each column multiplies all low or all high inputs together, so the outer columns are corner cases, not likely outcomes.

The output is not "$3,960 lost". It is a commercially meaningful, plausibly fixable gap whose exposure sits between $1,200 and $10,080 a month, which is enough to decide whether it earns a fix ahead of other work. If margin matters more than revenue, convert the same scenarios to contribution.

Why separate consideration risk from value-capture risk?

Because value can leak after the product wins.

A product loss means the AI recommended a rival's product. A merchant loss means it recommended yours but sent the shopper to another seller. In practice: Product A fits a high-demand need but rarely enters the shortlist; Product B is recommended often but the shopper goes to a marketplace. OpenAI treats seller choice as its own step, ranking a product's merchants on factors such as availability, price and quality (OpenAI Help Center).

Both cost money, but they have different causes and owners, so never add them into one "AI visibility loss". The model above handles the first through the participation gain and the second through preferred-store capture; our page on what seller choice does to margin covers the second in depth.

Gravton's free audit starts with a Demand Snapshot for your category: the questions your eligible products should win, where they go missing, who takes those answers and which gaps to fix first. Product & Retail Intelligence shows which products, retailers and sources AI recommends, and why.

Get your free AI visibility audit and see which of your gaps are worth fixing first.

See how AI shops your category.

Map demand, products considered, sale routing and the first gaps your team should fix.

See how AI shops your category.

Map demand, products considered, sale routing and the first gaps your team should fix.