After the Sale, AI Sends Customers Back to Your Website. Until Something Breaks

Most AI visibility work stops at checkout. Brands track whether ChatGPT recommends them, how they compare with competitors, which sources shape the shortlist. Then the customer buys, and the tracking ends. The customer's questions do not. How long do these shoes take to break in? Why is my email sync not working? How do I register my vacuum for warranty? More and more of these questions go to an AI assistant before they ever reach a help center or a support agent. The answers shape how customers use the product, whether they get unstuck, and whether they come back. The economics are well known. Research by Frederick Reichheld of Bain & Company found that increasing customer retention by 5% can increase profits by 25% to 95%, and Harvard Business Review  notes that, depending on the study and the industry, acquiring a new customer costs five to 25 times more than keeping one. So we asked a simple question: after the sale, who does AI send your customers to? We ran a controlled benchmark of matched questions asked before and after a purchase, across four product categories and five AI engines, read in their public interfaces. The answer is reassuring, with one important exception.

What we found

• After the sale, AI turns to the brand. 90% of post-purchase answers with sources cite the brand's official website, versus 17% before the purchase. The official site is the first source in 66% of post-purchase answers, versus 6% before.
• It is a different web. Official sites go from 3% to 50% of cited domains. Only 14% to 29% of the domains cited after the purchase were already cited before it.
• Help content does the work. 89% of the official pages cited after the purchase are help, support, warranty or account pages.
• Troubleshooting is the weak spot. For returns, warranty and support questions, every answer cites the brand. For troubleshooting, only 70% do, and the official site comes first in just 39% of answers.
• The pattern holds. In a repeat collection two days later, 93% of post-purchase answers cited the official site, and troubleshooting was again the weakest moment.
• The conversation keeps going. 68% of post-purchase answers end with an invitation to continue, and 18% of those bring commerce back: replacement parts, deals, protection plans.

Before the sale, the brand is one voice among many

Before a purchase, AI answers are built from third parties. Media, review sites and comparison pages account for 86% of the cited domains. The brand's own website appears in only 17% of answers, and it is the first source in just 6%. B2B software is the partial exception. In the CRM category, 42% of pre-purchase answers cited the vendor's site, because software companies publish pricing pages, plan comparisons and "vs" pages that assistants use as evidence. In consumer categories, the figure fell to between 5% and 10%. Before the sale, AI trusts what others say about you.

After the sale, AI goes to the source

Once the question is about using the product, the picture flips. 90% of post-purchase answers with sources cite the brand's official site, and it is the first source in 66% of them. The shift holds in every category we tested: from 42% to 96% for CRM software, from 10% to 82% for running shoes, from 9% to 92% for smartphones and from 5% to 90% for cordless vacuums. Engines differ in degree, not in direction. After the purchase, ChatGPT and Google AI Mode cited the official site in every answer with sources, Perplexity in 95%, Gemini in 79% and Copilot in 74%. The behavior starts before the answer is written. When building post-purchase answers, every one of ChatGPT's background searches named the brand, versus 63% before the purchase. In three answers out of four, at least one search used a site: operator pointing at a specific domain, a search pattern we documented in our analysis of ChatGPT's collapse in Reddit citations . What AI cites on your site also changes. Before the purchase, the few official pages cited were mostly product, pricing and comparison pages: only 13% were help content. After the purchase, 89% of the official pages cited were help centers, support articles, warranty pages, manuals and account settings.

A different web after the purchase

The sources that sold the product are not the ones that support it. Only 14% to 29% of the domains cited after a purchase were already cited before it, depending on the category. Review sites and comparison pages drop from 86% to 38% of cited domains. Retailers almost disappear, from 7% to 1%. Communities such as Reddit, YouTube and independent forums grow from 4% to 11%. This has a practical consequence. Your pre-purchase visibility tells you very little about your post-purchase visibility. A brand can dominate the "best of" lists and still lose its own customers to a third-party tutorial the day after delivery.

Troubleshooting is where brands lose the answer

Splitting post-purchase questions by customer moment reveals where the risk is:

• Returns, warranty, cancellation and support. Every answer cited the brand, and the official site came first in 96% of them. Policies only exist on the brand's site, and assistants know it.
• Care and maintenance. 96% cited the brand, 63% as the first source.
• Getting started. 95% cited the brand, 63% as the first source.
• Troubleshooting. Only 70% cited the brand, and just 39% as the first source. Media and review sites accounted for 47% of the cited domains, and communities for 13%.

The repeat collection confirmed the gap. Troubleshooting was again the only moment where the brand was missing from a significant share of answers (76% cited it, versus 95% to 100% for the other moments), and the one where it came first least often (57%, versus 76% to 84%). Troubleshooting questions are written in the customer's words, around a symptom: "Why does my Dyson V15 keep pulsing?", "Why is my Samsung Galaxy S25 battery draining fast?", "Why is my HubSpot email sync not working?". Help centers are usually organized around features and settings. When no official page matches the symptom, assistants assemble the answer from tech media, YouTube tutorials and forum threads.

This is also where customers struggle most. A Gartner survey of 5,728 customers  found that only 14% of customer service issues are fully resolved in self-service, while 73% of customers use self-service at some point in their customer service journey. Even issues customers describe as very simple are fully resolved only 36% of the time. AI assistants are becoming that self-service layer. If your troubleshooting content is not the one they cite, someone else's answer becomes your customer experience.

The post-purchase journey, moment by moment

The conversation doesn't end at the fix

Post-purchase answers rarely close the conversation. 68% of them end with an invitation to continue, and 95% of those invitations offer more help: a diagnosis, a checklist, the next step. More than half ask for specifics before going further, most often the exact model, version or purchase channel: which Ghost generation, whether the battery has a release button or three screws, whether the shoes were bought on the brand's site or from a retailer. Assistants are trying to match a precise situation. Generic help pages lose that match to sources that are specific. Commerce comes back too. 18% of post-purchase invitations bring up replacement batteries, current deals or extended protection plans. After the sale, AI is also deciding which accessories, upgrades and renewals your customers see. We explore how these follow-up turns shape decisions in a companion study: The First Answer Isn't the Decision .

A post-purchase playbook for AI visibility

1. Write troubleshooting content around symptoms, not features. Title articles the way customers describe the problem ("keeps pulsing", "battery draining fast", "email not syncing"), then give the diagnosis and the fix.
2. Be version-specific. Assistants ask which model, generation or plan the customer has. Publish answers that name versions explicitly, and keep older versions documented.
3. Keep one canonical page per policy. Returns, warranty registration, cancellation and support contacts should each have a single, current, crawlable page. AI already goes there first: make sure it finds the right one.
4. Make your help center readable by AI. Help content behind logins, rendered only in JavaScript or blocked in robots.txt cannot be cited. Check your robots.txt rules for AI crawlers  and how your pages render for AI crawlers .
5. Watch the communities that answer for you. In troubleshooting, forums and videos fill the gaps. The threads AI cites show which problems your official content does not yet cover.
6. Track post-purchase prompts as their own stage. Measure, per engine, how often your site is cited and whether it comes first, moment by moment.

How Qwairy helps

Qwairy reads answers in each engine's real interface, country by country, with their sources and links:

• The Post-purchase stage, added in Qwairy v1.20 , covers the questions people ask once they are customers. Assign it when you create or import a prompt, or pick it in Generate with AI. New workspaces start with a first Post-purchase prompt.
• Stage-level analysis lets you filter and compare results by stage across the dashboard, the GEO Matrix, exports, shared views, the REST API, MCP and the n8n node. You see whether you lead before the sale and fade after it.
• Sources show which domains AI cites for each prompt, so you can spot the tutorials, media and forums answering your customers' questions instead of you.
• Follow-Ups reveal the questions engines suggest next, including after the sale.

Methodology

• Questions. Matched questions in four categories: running shoes, CRM software, smartphones and cordless vacuums. Pre-purchase questions cover best options, head-to-head comparisons, value for money and specific needs. Post-purchase questions name a widely known product and cover four moments: getting started, care and maintenance, troubleshooting, and policies and support (returns, warranty, cancellation, contact). The brands named were chosen as familiar examples; their presence implies no relationship with Qwairy.
• Engines. ChatGPT (with search), Google Gemini, Google AI Mode, Microsoft Copilot and Perplexity, collected from their public web interfaces in the US, without login or personalization, on October 5, 2026, then again on October 7, 2026. One answer per question and engine in each collection; 97% and 98% of requests returned a usable answer. Figures in this article come from the first collection; the repeat collection checks their stability (for example, 90% then 93% of post-purchase answers cited the official site).
• Sources. Cited URLs were deduplicated by domain within each answer. Domains were classified as official (the brand's domains and subdomains, including brand-hosted help centers and communities), communities (Reddit, Quora, YouTube, forums and social platforms), retailers and marketplaces, or media, reviews and comparisons (all other domains). "First source" is the first cited domain in the answer. Technical redirect hosts, such as search API endpoints, were excluded. Help pages were identified from URL paths such as support, help, knowledge, warranty, manual or account.
• Limits. This is a controlled benchmark, not a measure of traffic. Answers vary between runs, accounts and countries, and the brands studied are large, well-documented companies: smaller brands may get cited less often after the sale. Percentages are rounded.