Run AEO like a performance channel: the answer-share scorecard for D2C brands

Performance marketers already know how to run a channel. There is an account structure, a handful of numbers everyone trusts, rules that fire when something breaks, and a weekly review where decisions get made. Most D2C brands run AEO with none of that. Someone checks ChatGPT, screenshots an answer, ships a blog post and checks again a month later.

That does not hold up, because AI answers move every week. Engines change which sources they trust, competitors publish new comparison pages, and your own catalogue shifts with launches, price changes and stock-outs. A fix that won an answer in March can be gone by May, and without a steady measure you will not know why.

This is the system we use at Eldor to run AEO, or GEO if that is the term your team uses, like any other performance channel. It has four parts: a prompt portfolio that works like an account structure, a four-number scorecard we call answer share, a rule for deciding what to fix first, and a weekly rhythm that keeps it moving.

Part 1: build a prompt portfolio, not a prompt list

Your prompts are your keywords. A random list of questions gives you random numbers, so structure them the way you would structure campaigns. We group every prompt into one of four buckets, because each bucket fails for different reasons and needs different fixes:

• Category prompts. The shopper knows the need, not the brand. "Best vitamin C serum for oily skin under ₹1,000." This is where new customers come from, and usually where a challenger brand is weakest.

• Comparison prompts. The shopper is choosing between names. "[Your brand] vs [competitor]", "alternatives to [competitor]". These are late-funnel and high intent.

• Problem and use-case prompts. The shopper describes a situation. "Serum that won't pill under sunscreen", "gift for a runner under ₹3,000". These reward brands with specific, well-described products.

• Brand and policy prompts. The shopper already knows you. "Is [brand] cruelty-free?", "[brand] return policy", "does [brand] offer COD?". You should win almost all of these. When you don't, something is broken.

The best prompts come from places a performance team already has access to:

• Paid search term reports. The long, question-shaped queries in your Google Ads search terms report are the closest thing to real AI prompts you will find.

• Search Console queries that start with how, which, best, vs or is.

• Site search logs, which show what shoppers could not find in your navigation.

• Support tickets and WhatsApp chats. Every repeated pre-purchase question is a prompt someone is also asking an AI engine.

• Reviews and marketplace Q and A on Amazon, Nykaa, Myntra or Flipkart, written in your customers' own words.

Tag every prompt with its bucket, the product line it maps to, the price band and the persona. Then weight prompts the way you weight budget: a prompt that maps to your best-selling product line matters more than one that maps to a slow seller. Start with 25 to 50 prompts, and grow toward 100 or more once the weekly routine is running.

Part 2: the answer-share scorecard

"Are we showing up in ChatGPT?" is a yes or no question, and it is the wrong one. We track four numbers, and we call them the four Ps of answer share. Not those four Ps. Each one maps to a metric a performance marketer already uses.

1. Presence: are you named at all?

The share of tracked prompt runs in which your brand or product is named or cited. If you track 100 prompts on three engines, that is 300 answers a week, and presence is the share that include you. It is the AEO version of impression share: you cannot win an auction you are not in.

2. Prominence: how are you named?

Being listed seventh in a paragraph is not the same as being the top pick. For every answer you appear in, score how you appear: 3 if you are the first or recommended pick, 2 if you are one of the listed options, 1 if you get a passing mention. Prominence is the average of those scores. Think of it as top-of-page rate. A rising presence with falling prominence means engines know you but do not prefer you.

3. Precision: is what it says about you true?

The share of answers that mention you in which your product facts match your live store: price, availability, variants, key claims and policies. This is the number most AEO reports leave out, and for a D2C brand it is often the most expensive. An answer that recommends you at last season's price, or says a bestseller is out of stock, sends a shopper away already disappointed. The closest ad metric is message match: what the shopper was told has to match what they land on.

4. Pull: does it send shoppers who buy?

AI referral sessions, add-to-carts and orders. This is clicks and conversions, and it is the hardest of the four to measure, so set it up carefully:

• In GA4, create a custom channel group for AI referrals and place it above Referral. Match session sources such as chatgpt.com, perplexity.ai, gemini.google.com, copilot.microsoft.com and claude.ai.

• ChatGPT adds utm_source=chatgpt.com to some outbound links, but not all of them. Many AI visits arrive with no referrer and land in Direct.

• Google counts AI Overviews and AI Mode traffic inside Search Console's normal web totals, so you cannot cleanly separate it from organic search.

Treat the pull you can see as a floor, not the full number. Its job is to show the trend and to tie answer share to revenue, not to settle attribution arguments.

Report all four numbers by bucket and by engine, and judge them on a four-week rolling trend. AI answers vary from run to run, so a single week's movement is noise until it repeats.

Part 3: fix the weakest number first

The four numbers form a chain. A shopper cannot see an inaccurate answer about you if you are never named, and a well-placed answer cannot pull revenue if it quotes the wrong price. So each week, look at each bucket and find the first number in the chain that is weak. That number tells you what kind of problem you have, and what kind of work fixes it.

• Presence is weak: a coverage problem. Engines are not finding you for these prompts. You may have no page shaped like the question, the page may not be reachable by AI crawlers, or you may be missing from the third-party sources engines cite for this category, such as listicles, marketplace pages and forums. First fixes: build the answer-shaped page (a buying guide, a comparison, a collection organised by need), check crawler access, and look at which sources get cited so you know where you need to appear. Our guide to how answer engines find, choose and quote your products covers this in detail.

• Presence is fine, prominence is weak: an evidence problem. Engines know you but prefer someone else. Usually a competitor gives them clearer reasons: specific claims, proof, and a stated "best for". First fixes: state what each hero product is best for, add concrete evidence (ingredient percentages, certifications, review counts and what reviewers say), publish honest comparison pages, and earn mentions in the sources engines already trust.

• Prominence is fine, precision is weak: a data problem. You are recommended, but with the wrong facts. The cause is usually stale or scattered product data: an old price on a reseller listing, a feed that has not synced, or a policy that only lives in a PDF. First fixes: correct the product data at the source, keep your feed and structured data in sync with the store, and update third-party listings that engines are reading.

• Precision is fine, pull is weak: a landing problem. Answers are good but shoppers do not convert. Check where the cited links go. A citation that lands on the homepage, a product that is out of stock, a slow page or a page with no clear next step loses the shopper you just won. First fixes: make the pages engines cite ready to convert, with stock, price and the next step clear above the fold.

This rule stops the most common waste in AEO: writing more content to fix a problem that is really about data, or polishing product pages for prompts where you are not even present.

Part 4: set rules before you need them

Ad accounts run on automated rules because nobody can watch every campaign every day. AEO needs the same thing. Write down your rules once, then let the weekly review (or your tooling) check them. Here are the ones we would start with, with thresholds you should tune to your own numbers:

• If presence in a bucket falls for two weeks in a row, open the sources now being cited for those prompts. Someone new is usually in them.

• If a competitor replaces you on a prompt you held for a month, compare their cited page with yours, claim by claim.

• If any answer quotes the wrong price or stock for a top-20 product, fix it within the week. Precision misses on hero products are the most expensive kind.

• When you launch a product, add three to five prompts for it on launch day, so you have its starting point.

• When you change prices or run a sale, re-check precision for those products in the next run.

• If pull falls while presence holds, check the landing pages before touching content.

Part 5: the weekly rhythm

The system only works if it runs every week, including the weeks when everyone is busy with a sale. We keep it to a rhythm small enough to survive a launch calendar:

• Monday: read the scorecard. Thirty minutes, the same way you would review paid media. Look at each bucket, find the weakest number, and pick one to three fixes for the week. No more.

• Tuesday to Thursday: ship. Make the fixes the weakest numbers point to. One well-chosen fix that ships beats five that sit in a doc.

• Friday: log the change. Write down the date, the page or data you changed, and the prompts it targets. When a number moves in three weeks, the log tells you why. This is the step most teams skip, and the reason most teams cannot say what worked.

• Monthly: rebalance the portfolio. Add prompts from new search terms, support questions and launches. Retire prompts nobody asks any more, and re-weight by revenue.

• Quarterly: reset the baseline. Re-run the full portfolio, compare it with the last quarter and set targets for the next one.

Mistakes we see most often

• Judging on a single answer. One screenshot is one sample. Decide on four-week trends.

• Tracking only brand prompts. You will look great and learn nothing. Category and comparison prompts are where share is won.

• Reporting presence without precision. A rising mention count can hide answers that quote the wrong price.

• Changing ten things at once. Without a change log, you cannot tell which fix moved the number.

• Measuring AI traffic only from referrals. Visible AI referrals undercount the real number, so use them for trend, not totals.

Where Eldor fits

The free Eldor AI audit gives you a starting scorecard: 25 shopper prompts written for your category and price band on ChatGPT and Google AI Overviews, and 100 of your products checked against your live store, with the competitor and source behind every lost answer.

On a paid plan, Eldor runs the weekly part for you. It tracks 100 or 300 prompts every week across your engines, shows share of voice and the sources each engine cites, flags every answer that quotes the wrong price, stock or policy, and fixes product data at the source with enrichment and page rewrites. Your team keeps the Monday review and the decisions. Eldor keeps the numbers current.

FAQ

What is answer share?

Answer share is how much of the AI answer space in your category you hold. We measure it with four numbers: presence (are you named), prominence (how you are named), precision (are your facts right) and pull (do those answers send shoppers who buy).

Is AEO the same as GEO?

In practice, yes. Answer engine optimisation and generative engine optimisation both describe improving how a brand appears in AI-generated answers. Teams use the terms interchangeably.

How many prompts should a D2C brand track?

Start with 25 to 50 across the four buckets, weighted toward category and comparison prompts. Grow toward 100 or more once the weekly review is a habit, adding prompts for every product line that matters to revenue.

How often should we measure?

Weekly. AI answers vary from run to run, so measure on a fixed weekly schedule and judge changes on a four-week rolling trend rather than a single result.

How do we track AI traffic in GA4?

Create a custom channel group for AI referrals that matches sources such as chatgpt.com, perplexity.ai, gemini.google.com, copilot.microsoft.com and claude.ai, and place it above Referral. Expect it to undercount, because many AI visits arrive without a referrer and Google's AI features are counted inside normal search traffic.