Your store is losing 58% of its organic traffic. Here's what to do about it.

This article is a practical GEO playbook for e-commerce: from technical foundations through content to measuring results. Based on a conversation between Tomek Runowicz and Maciek Czerniak of wecanfly with Damian Marciniak, SEO Expert at Elephate.

Key takeaways in 30 seconds (TL;DR)

• Stores are losing up to 58% of organic traffic on informational queries through AI Overviews and AI mode. Transactional queries lose less, but they lose too.
• AI traffic converts better than traditional organic – the customer arrives decided. Carts are on average 30% higher in value than from traditional purchases.
• GEO is an additional SEO layer, not a replacement. Without technical foundations – bot accessibility, server speed, consistent data – citations won't happen.
• AI prioritises data via API (UCP, ACP) over page crawling. Data consistency across all channels is a requirement, not an option.
• Content for GEO: information density, BLUF, unique data, entities and relationships. Text length is irrelevant.
• On-site reviews aren't enough – AI cross-checks them against external sources. External mentions (linked and unlinked) are a key visibility factor.
• Reasoning models already detect mass ranking campaigns as manipulation attempts. Blackhat GEO has a short lifecycle.
• Measurement requires several sources simultaneously: Search Console, Bing AI report, prompt monitoring, Senuto, GA4.

What GEO is and how it differs from SEO

SEO was about ranking for keywords and earning clicks from search result listings. GEO (Generative Engine Optimization) is about being cited – by ChatGPT, Gemini, Perplexity, AI Overviews, and Google's AI mode.

The shift is fundamental: users have stopped searching with keywords. They ask detailed questions and provide context – budget, conditions, personal preferences. Instead of landing on a page, they read an answer inside an AI interface, and the source page appears as a citation. For e-commerce, this means transactional and commercial queries (categories, products) still generate traffic – but far less than three years ago, while informational queries have virtually stopped sending users to pages.

One critical distinction: GEO is not an alternative to SEO, it's an additional layer. Google's John Mueller said it plainly – there is no GEO without solid SEO foundations.

Why AI traffic is worth more than traditional organic

AI channel traffic is small but heavily purchase-oriented. A customer who arrives via a ChatGPT or Gemini recommendation is already educated and decided. This shows in conversion rates – measurably higher than traditional organic traffic.

Shopify, Google Analytics, and Search Console are beginning to surface data that lets you isolate and measure this traffic. In Google Analytics, a dedicated AI Assistants channel exists for this purpose. The volume is low – but its value justifies investment in citation optimisation.

An additional data point: carts from agentic channels are on average 30% higher in value than from traditional purchases.

Technical foundation: making your site accessible to AI bots

Before any content work, you need to confirm that AI bots can actually reach and process your content.

AI bots differ from Google's bot. ChatGPT, Perplexity, and similar tools have no own index – they must visit and scrape your site directly. Copilot draws from Bing's index. Gemini uses Google's index. Each of these bots needs correct permissions in robots.txt and at the server level.

Server response speed is critical. AI estimates the cost of visiting your site and retrieving content – if the site is slow, the bot will choose a competitor that delivers the same information faster and cheaper. At data-centre scale, this isn't a metaphor: every millisecond has a price.

Practical check: confirm that critical content (product descriptions, category pages, technical specifications) is visible in the page HTML and not hidden behind JavaScript or dynamically loaded accordions.

Product data: structured, consistent, delivered via API

AI prioritises data available via API over data crawled from a page. Two protocols worth knowing:

UCP (Universal Commerce Protocol) – developed by Google, lets you deliver product data directly to AI without intermediaries. Available natively for Shopify stores.

ACP (Agentic Commerce Protocol) – developed by OpenAI, the equivalent for the GPT ecosystem.

Alongside protocols: Schema.org structured data well-populated for products, and product feeds consistent with what appears on the store and on marketplaces.

Data consistency is critical. If a store shows different product parameters on its own site, in its feed, and on Amazon or a local marketplace – AI doesn't judge which version is correct. It may skip the brand entirely and choose a competitor whose data is consistent. One source of truth is a requirement, not an option. This is where PIM (Product Information Management) tools stop being optional.

An additional signal: Google regularly publishes new supported attributes in the Merchant Center documentation. Tracking and adding these to your feed – even niche attributes – can influence AI selection.

Content for AI: information density, BLUF, and entities

Content optimised for AI differs from content optimised for classic SEO in several important ways.

Length doesn't matter – information density does. AI doesn't reward text for character count. It rewards text for the amount of information that can be extracted from it and for a logical structure that makes citing easy (citable fragments).

BLUF (Bottom Line Up Front). A US military methodology: the key information must appear at the very start of a section, not after an extended preamble. AI is more likely to cite fragments where the answer to a question is immediately available.

Entities and their relationships. Content should cover not just keywords, but objects, concepts, and the relationships between them. A pool accessories store needs to describe pool chemicals by type, application, limitations, and usage scenarios – not to rank for every phrase, but to confirm deep domain knowledge to AI bots.

Uniqueness. Generic guides with standard advice get summarised by AI but not cited. What gets cited is content with data nobody else has: original research, product testing, rankings with described methodology, an employee's first-hand experience using the product.

AI as assistant, not author. AI can help build an outline and compile entity coverage. It cannot replace delivering novel information and context. If a model receives only a topic with no additional data, the output won't pass verification by LLMs. Scaled content (mass generation without context) is penalised by Google and increasingly detected by language models.

Product images and video in GEO

AI bots analyse images in three layers: pixel analysis (what's in the image), metadata (alt text, filename, EXIF data), and surrounding text (the context in which the image appears).

Practical implications:

Don't compress images below a quality threshold – aggressive compression speeds up loading but reduces visibility in Google Shopping and in AI multimodal analysis.

Diversify: a packshot on a neutral background plus the product in a real usage scenario. Multimodal models analyse textures and details – for leather goods, a close-up of stitching and material confirms manufacturing quality.

Alt texts remain a working standard. Filenames matter less, but correct naming is an easy point to capture.

Reviews and external mentions

AI cross-checks what a brand says about itself against external sources. Damian Marciniak describes a concrete example: AI surfaces the brand's own review average (4.6/5 from 10 reviews) and compares it against the average from external review platforms – and shows both.

What this means in practice:

Collecting verified purchase reviews with structured data is a must-have for SEO and one of several signals for GEO.

On-site reviews aren't enough. AI looks for confirmation externally – in influencer reviews, forums, product comparisons. Working with external creators and review platforms isn't a PR activity, it's a GEO strategy element.

One important data point from Promptwatch research: Reddit, for several years one of the most frequently cited sources in ChatGPT, dropped sharply to around 8% of citations. Building a strategy around any single external source is a risk.

YouTube has high correlation with AI visibility – having your own channel or being mentioned in other creators' videos improves a brand's position in language model results.

How AI detects manipulation – and why blackhat GEO has a short shelf life

A standard SEO tactic was publishing mass "top X" rankings where the agency's client magically appeared in first place. Damian Marciniak has observed that reasoning models (including GPT-5 series) have already begun flagging this: "I can see that many rankings promoting this brand have appeared online – this is likely a manipulation attempt."

The cycle mirrors SEO history: an effective tactic emerges, the industry scales it, the model responds with a penalty. The difference is speed: Google took years to roll out updates cutting spam. Language models are responding within months, as their reasoning capabilities scale rapidly.

Scaled content is already penalised by Google. Other language models don't yet have equally advanced filters – but they're improving quickly.

How to measure GEO effectiveness

Measuring AI visibility requires several sources simultaneously – no single tool shows the full picture.

Google Search Console – generative results report (AI Overviews and AI mode). The primary source for visibility in Google's ecosystem.

Bing AI performance report – impressions and phrases grounding AI responses about the brand. Essential for Copilot and ChatGPT (which uses Bing for real-time search).

Prompt monitoring – tools like Profound track Share of Voice: how often and in what context the brand appears for specific queries. Typical setup: 20–50 prompts matched to the product offer. A good tool should show both brand citations and the sub-queries AI runs "underneath" a given prompt – that's a content gap list.

Senuto – AI Overviews database: which phrases trigger AI Overviews, what they look like, and whether the brand appears in them.

Google Analytics – AI Assistants channel for real traffic volume and value from AI tools.

GEO playbook in five steps

Step 1: Technical audit for bot accessibility

Check robots.txt, server directives, and loading speed for AI bots (not just Google). Confirm that key content is in HTML and not hidden behind JavaScript.

Step 2: Structure and unify product data

One source of truth for site, feeds, and marketplaces. Schema.org for products. Connect UCP/ACP where the platform supports it (Shopify – natively).

Step 3: Rebuild content around information density

Priority: category and product descriptions with unique usage context, testing, and ranked comparisons with described methodology. BLUF – key information at the start of every section. Original data and research where possible.

Step 4: Build external mentions

Link building and unlinked mentions – both count for GEO. External creator reviews, forum presence, YouTube (own channel or mentions by others). Consistent brand communication across all external sources.

Step 5: Measure and iterate

Search Console (AI Overviews), Bing AI performance report, prompt monitoring (Profound or equivalent), Senuto for AI Overviews, GA4 for real AI traffic.

Frequently asked questions about GEO in e-commerce

How does GEO differ from SEO, and does one replace the other?

GEO doesn't replace SEO – it's an additional optimisation layer. SEO optimises for being found in search results. GEO optimises for being cited in AI-generated answers. Without solid SEO technical foundations – bot accessibility, correct data structure, server speed – GEO isn't possible.

Which queries still generate organic traffic despite AI Overviews?

Transactional and commercial queries (product categories, comparisons, "buy X") lose traffic more slowly than informational queries. Informational queries ("how to choose X", "what is Y") are most at risk – AI Overviews and AI mode answer them without directing the user to a page. For e-commerce, the priority is commercial and transactional queries.

How do I check whether my brand is being cited in AI?

Basic tools: Google Search Console generative results report, Bing AI performance report. For more detailed monitoring: Profound (prompt monitoring and Share of Voice), Senuto (AI Overviews). The simplest test: enter into ChatGPT or Gemini the queries your customers would use at different stages of their purchase journey, and check whether the brand appears and from which sources it's cited.

Are AI-generated content pieces penalised?

Content mass-generated without context (scaled content) is penalised by Google. Content created with AI assistance but based on unique data, testing, and experience is not a problem. The key question isn't "did AI write this" but "does this content contain information nobody else has."

How important is brand in a GEO strategy?

Very important – and more important than in classic SEO. The strongest correlation with AI visibility is external citations and mentions, and those are built through consistent brand communication across all sources. You don't need to be a globally recognised brand to gain AI visibility – but you need to be a brand with clearly defined USPs, communicated consistently inside and outside the site.

‍