Start before someone searches for your product
OpenAI’s September 2025 study of 1.5 million consumer conversations grouped 49% of messages as asking, 40% as doing, and 11% as expressing. Around 70% of consumer usage was outside work. That doesn’t make half the messages shopping searches. It shows how much people use the product to ask for information and advice.
Think about what your buyer needs to understand before choosing a supplier. What problem do they have? Which approaches are worth comparing? What could go wrong? I’d start there, then move towards pricing, alternatives, and your specific offer. If all your content starts with your company name, you may be arriving late to the conversation.
Read the answer, even when it doesn’t send a click
Pew’s study of browsing from 900 US adults found traditional-result clicks in 8% of Google visits with AI summaries, compared with 15% without. Source links in the summaries were clicked in 1%. The browsing came from March 2025 and the results were reconstructed in April, so this is a dated observational study.
For your strategy, I’d check what the answer teaches someone before worrying only about the link. Does it describe the problem fairly? Is your product relevant? Are your limitations explained? Website traffic still matters, but it won’t show everything a buyer learned before getting there.
Look at the quality of the visitors who do arrive
Adobe’s 29 October 2026 retail report says AI-referred visits in September converted to purchases 90% better than non-AI visits and generated 83% higher revenue per visit. AI referral traffic was up 71% year over year. The analysis covered more than one trillion US retail visits, 100 million products, and 18 categories.
Those are relative comparisons among recorded visits, not a 90% conversion rate or a forecast for your business. Still, they give you a reason to ask whether the people arriving from AI have already narrowed their choices. Check inquiry quality and purchase behaviour as well as visit count. A smaller stream can deserve attention.
Decide which business problem you’re solving
An AI answer, a website visit, and a purchase are different moments. Fewer people clicking can coexist with stronger conversion among those who do. The studies above cover different groups and periods, so you can’t multiply their figures into your own forecast.
I’d ask your team a practical question: do you need more people to consider you, better information about the offer, or more suitable people reaching sales? Each problem suggests different work. A citation chart helps you investigate exposure; it doesn’t decide which problem matters most to the business.
| Observed unit | What it establishes | What remains unknown |
|---|---|---|
| The name appeared in a sampled answer | How many buyers saw or trusted it | |
| A page was linked as evidence | Whether the company was recommended | |
| Someone reached a measured website | Whether the visit was additional | |
| A commercial outcome occurred | What would have happened without the exposure |
Write something the answer can actually use
The GEO paper accepted at KDD 2024 tested a benchmark of 10,000 queries and reported visibility improvements of up to 40% under its experimental measures. It looked at changes such as adding relevant sources, quotations, and statistics. The effect varied by domain and by how visibility was measured.
The useful lesson is to improve the substance and clarity of the page. Give a real example, explain a condition, show the source for a claim, or answer the comparison the buyer is making. Don’t use the paper’s 40% as a lead forecast or fill a page with statistics just to imitate the treatment.
Keep the SEO basics in the plan
Google’s guidance explains that AI answers can use its Search index and related searches through query fan-out. It still recommends useful, original information and normal search fundamentals. You don’t need a separate page for every possible wording of the same question.
I’d organise pages around the information a buyer needs: pricing, compatibility, implementation, risks, and proof. One answer may draw on several of those pages. Make each one clear enough to be useful on its own, and connect them so someone can continue their research. That’s a content strategy you can maintain.
Check that the right crawlers can reach you
OpenAI documents OAI-SearchBot for search and GPTBot separately for model training. Your policy on training can differ from your policy on appearing in search. Ask whoever manages the site to check the actual rules and access, rather than assuming all AI crawlers do the same job.
When a page is absent from an answer, first rule out an access problem. A blocked page and a readable page that wasn’t selected need different fixes. Being allowed to crawl is only the starting point; it doesn’t guarantee a citation or recommendation.
Review the platform settings too
Google’s Search Console control for generative AI features rolled out worldwide on 31 August 2026. Inclusion is the default, but a property can inherit a parent property’s choice. Google distinguishes this setting from training controls and from exclusion from Search overall.
I’d check the relevant property with the site owner before commissioning more content. An inherited exclusion, a crawl block, and a weak explanation can all leave you missing from an answer. You’ll save time by finding out which issue you actually have.
Use Google’s reporting for the question it answers
Google’s dedicated Generative AI performance report gives a view of impressions for supported Search AI features, grouped by page, country, date, and device. It rolled out worldwide on 31 August 2026, and sites with too few impressions may not see it. Use it to check link exposure. Pair it with saved answers, visits, and inquiries when you want to understand the wider journey.
Use Bing to find where your sources participate
Bing’s February 2026 AI Performance announcement describes citation counts, cited URLs, sampled grounding queries, and trends across supported experiences. It separates those figures from a page’s rank or placement in an answer. I’d use the report to find which pages contribute, then open the answers to understand their role. A technical source and a buying guide may each earn a citation while helping very different decisions.
Turn the research into a plan you can run
Here’s how I’d start: choose a buyer and market, write down their important decisions, save the current answers, and inspect the sources. Prioritise gaps your team can fix. Give each change a clear owner and review date, then recheck the same questions. Track visits and suitable inquiries where you can.
At Searchable, I can help you build a custom Prompt Universe and get the team working through those findings. Bring your actual offer and the questions you hear in sales. If you want to see how that approach changes for a local service, read the senior living guide next: care fit, geography, and family expectations make the details especially important.
How to read the research
The sources were reviewed on 3 November 2026. These studies cover different years, products, and populations. I’ve kept their figures attributed and separated my recommendations from their findings. None establishes a standard revenue uplift for every business. Use them to ask better questions about your own market.
Sources and details
Sources reviewed: November 3, 2026. I work at Searchable. The examples and recommendations are mine; reported results are credited to their sources.
1
Searchable