Does AI Cite E-commerce Sites More the Closer Users Get to Buying?

Two pieces of advice come up often in GEO content planning: the closer users are to placing an order, the more you should invest in e-commerce and review content; and since AI cites media often, get more media coverage. Both assume that every industry follows the same path to purchase.

The Dageno team compiled answers from 5 AI platforms in the Dageno AI answer database from June to September 2026: ChatGPT, Google AI Mode, Google AI Overviews, Gemini and Copilot. Each question carries a buying-stage tag: awareness means "understand it first", consideration means "compare options", and decision means "confirm the choice or the purchase terms". We also looked separately at 4 industry question sets: charging and power banks, AI gateways (services that give one access point to several AI models), omnichannel customer messaging (tools that bring customer messages from every channel together) and web scraping APIs (services that let programs extract data from web pages).

1. The closer users get to buying, the more often do AI answers mention a brand name?

Yes. This holds on all 5 platforms, and the gap is widest on ChatGPT.

In September, the share of answers that mentioned at least one brand name:

• ChatGPT: awareness 73.7%, consideration 88.5%, decision 92.6%. The gap between awareness and decision is 18.9 percentage points.
• Google AI Mode: 86.2%, 93.3%, 96.0%.
• Google AI Overviews: 83.5%, 92.6%, 94.3%.
• Gemini: 83.3%, 94.8%, 95.7%.
• Copilot: 82.1%, 93.8%, 95.9%.

On ChatGPT this pattern held in all 4 months: decision questions were between 92.6% and 93.8% every month, awareness questions between 69.0% and 73.7%. From June to August, however, only part of the answers carried stage tags, so those months are reference only.

This shows that the closer a question gets to a choice, the more likely AI is to name specific brands or products. For a brand, writing only "what is it" concept content may not reach the questions where comparison actually happens.

But more brand mentions do not mean this stage is easier for your brand to get picked. The figure counts "any brand appeared"; every brand named in an answer could be a competitor. A decision question asks "which one, given this budget, this scenario and these constraints", and that needs comparable conditions. If a competitor has compatibility notes, test methods and usage limits, and you only have "excellent performance", you are unlikely to be chosen (inference).

2. Is the conclusion the same when questions are grouped by search intent?

The direction is similar: transactional questions had the highest share of brand mentions on 3 platforms.

In September, 92.0% of ChatGPT answers to transactional questions mentioned a brand name, against 82.4% for informational, 84.7% for commercial and 84.4% for navigational questions. Google AI Mode and Gemini were also highest on transactional questions, at 95.9% and 95.3%. Google AI Overviews and Copilot were highest on navigational questions, at 96.4% and 97.9%. Navigational questions often look for a specific brand's website directly, so the brand name is already in the question (inference).

Search intent (informational, commercial, transactional, navigational) and buying stage (awareness, consideration, decision) are two separate sets of tags. They do not map one to one, and one question can carry several tags at once, so the two sets of numbers cannot be added together. Navigational questions in particular should not be mixed with brand-free choice questions into one "brand exposure" score.

3. Do questions closer to purchase cite e-commerce sites more often?

With all industries combined, yes.

The comparison here uses "coverage": out of every 100 answers with links, how many include at least one link to this type of site. One answer can link to a brand site, an e-commerce site and a media site at the same time, so the types add up to more than 100%.

In September, the share of answers with e-commerce links, awareness vs decision:

• Google AI Overviews: 9.6% → 29.7%, the widest gap.
• Copilot: 13.6% → 28.2%.
• Google AI Mode: 10.6% → 16.9%.
• ChatGPT: 15.3% → 19.2%.
• Gemini: 4.8% → 6.4%.

But this is not a straight line of "closer to the order, more e-commerce". On ChatGPT, consideration questions had an e-commerce share of 13.9%, lower than awareness questions. Brand sites did not rise either: on Google AI Mode, Gemini and Copilot, decision questions included brand-site links less often than awareness questions, at 92.3% vs 94.9%, 79.6% vs 84.4% and 72.0% vs 79.1%.

So close to purchase, brand sites, e-commerce, media and communities may all appear in the same answer. There is no fixed relay of "brand site first, then e-commerce".

4. Does this pattern still hold in a specific industry?

Not necessarily. In a different industry, the direction can be the opposite.

ChatGPT in September, share of answers with e-commerce links across the 4 industry question sets:

Table: share of answers with links that cite at least one e-commerce site. The decision groups have 88, 1,605, 84 and 265 answers with links respectively.

• Charging and power banks: 86.4% of decision answers included an e-commerce link, far above the 20.0% for awareness. When buying hardware, the last step is confirming where to buy and which version.
• Web scraping APIs: the closer to purchase, the lower the e-commerce share, from 8.5% down to 4.3%. In the 1,605 decision answers, brand sites were at 100%. When buying a technical service, the last thing to check is the solution itself.
• Omnichannel customer messaging: 35.5% of decision answers included a media link, against only 2.4% for consideration. That 35.5% comes entirely from YouTube, meaning that at the final confirmation stage, AI often cites product demos and tutorial videos.
• AI gateways: 28.3% of awareness answers included an e-commerce link, almost all from amazon.com. These are most likely AWS technical docs, because the amazon.com domain is classed as e-commerce as a whole (inference; the data only has domains, not individual pages).

Note that the decision groups for charging and power banks and for AI gateways have only 88 and 84 answers with links. Those two figures are leads worth checking, not a precise picture of the whole industry.

So "the closer to the order, the more e-commerce content" only fits some industries. A hardware brand's last step is checking that model numbers, accessories and pricing terms match across sales channels. A technical service has to explain integration requirements, API capabilities, failure handling and cost.

5. Which websites do these links actually come from?

Different for every stage and every industry. When site types cannot explain something, look at the specific sites.

ChatGPT's charging and power bank questions in September:

• Awareness: the review site chargetechlab.com covered 16.7% of answers with links, gearlark.com 10.4%. The US Federal Aviation Administration, faa.gov, was at 6.8%, because these questions often ask whether a power bank can go on a plane.
• Consideration: chargetechlab.com covered 24.2%, gearlark.com 12.7%, soldermag.com 11.5%. In the taxonomy, these review sites fall under "Other", not "Media".
• Decision: Walmart covered 73.9%, Target 43.2%, Amazon 36.4%, all retail platforms.

The other industries have their own profiles: in web scraping API awareness questions, the academic publisher springer.com covered 16.1%; in AI gateway awareness questions, the US National Institute of Standards and Technology, nist.gov, covered 20.3%.

These examples show that the same "awareness" stage means transport rules and reviews for power bank questions, and standards and academic material for technical services. Seeing NIST, you cannot treat it as a review outlet; seeing a commercial site among the sources, you cannot assume it is an independent third party. To decide what to learn from, you still have to open the cited page and see what it answers.

6. Do sources in different industries move in the same direction each month?

No.

Among ChatGPT answers with links, the share citing media and content platforms:

• Charging and power banks: July 35.8%, August 29.8%, September 20.8%.
• AI gateways: 9.5%, 7.8%, 3.6%.
• Omnichannel customer messaging: 5.3%, 3.9%, 2.4%.
• Web scraping APIs: 3.7%, 3.9%, 4.6% (June was 3.0%).

The media share for charging and power banks was far above the other 3 industries every month, but it also fell the fastest; web scraping APIs went up instead. So "AI has been citing media less lately" is not a premise that applies to every industry.

The e-commerce share behaves the same way. With all industries combined, from June to September, the share of ChatGPT answers with e-commerce links fell from 20.7% to 15.0% and Google AI Overviews from 24.3% to 15.3%, while Copilot rose from 19.8% to 21.8%; Google AI Mode went 15.1%, 27.6%, 26.6%, 11.6%, up and down.

Questions and collection volume differ by month, so these changes cannot be attributed to the platforms alone. What they show is that one source analysis across all industries cannot be used forever.

7. What should marketing teams do next?

1. Don't apply a generic purchase funnel. "More e-commerce content the closer to the order" only holds in some industries. Start with your own industry: separate awareness, consideration and decision questions, and look at which site types AI actually cites.
2. Prepare evidence for the judgment the user has to make, instead of spreading articles evenly across stages. Take power banks as an example (hypothetical): a user might first ask whether it can go on a plane, then compare charging capability, then confirm which model to buy. Those three judgments need, respectively, the original regulation and accurate specs; compatibility notes and test conditions that can be checked; and model and accessory information that matches between the brand site and the sales channels. For a web scraping service, the judgments are target sites, data formats, concurrency, failure rates and total cost.
3. Hardware brands: check your sales channels first. In charging and power bank decision questions, retail platforms covered most answers. Models, variants, accessories and pricing terms have to match between the brand site and the retail platforms.
4. Technical services: get the website clear first. In web scraping API decision questions, brand sites covered every answer with links, and e-commerce was under 5%. Integration requirements, API capabilities, limits and cost should be written on the website where they can be cited directly.
5. Look at specific websites, not just site types. Review sites may be classed as "Other", AWS docs as "E-commerce", and product demo videos as "Media". Before deciding on a partnership or new content, open the cited page.
6. Keep "a brand name appeared" and "you were chosen" separate. Many brand mentions in decision questions do not mean your brand is among them. Check who the answer names and on what grounds, then re-test monthly with the same set of questions.

Method

The data in this article comes from the Dageno AI answer database. From June to September 2026, we analyzed 10,045,768 answers from ChatGPT, Google AI Mode, Google AI Overviews, Gemini and Copilot. The questions come from Dageno's historical monitoring and cover different countries, languages and industries. It is a fixed question set, not a random sample of user queries.

Buying stage and search intent are tags that come with each question. In September, between 95.7% and 99.7% of answers on the 5 platforms carried a stage tag, 1,259,780 answers in total; coverage was lower from June to August (for example, only 39.2% for ChatGPT in June), so comparisons by stage use September only. One question can carry both an intent tag and a stage tag, so the groups cannot be added together.

The 4 industry question sets total 450,570 answers. Web scraping APIs cover June to September and all 5 platforms; the other 3 industries start in July and do not cover every platform: charging and power banks has ChatGPT, Google AI Mode and Gemini, AI gateways has ChatGPT and Google AI Mode, and omnichannel customer messaging has ChatGPT and Google AI Overviews. The question mix and regional make-up differ between industries, so the comparison is meant to surface differences, not to rank industries.

"Brand mention" uses all answers in the group as the denominator and is detected automatically; coverage uses answers with links in the group as the denominator and follows Dageno's existing website taxonomy. A link appearing in an answer does not mean the answer text used its content, nor does it mean a recommendation, a click or a purchase.