Goal
How visible are you when the search result already contains the answer?
Search is spreading across more surfaces than two years ago. The work behind it has stayed the same: be reachable, answer one question unambiguously, and be able to back the answer up.
What we do: We check what Google finds and processes, pick up the demand already in your performance report and measure on a fixed schedule whether AI answers name you; you are visible where search delivers the answer itself.
Why this is hard right now
There is no Search Console for visibility in AI answers. The answer only becomes reliable when the same questions are asked regularly and in many phrasings, with the cited sources recorded. Because answers fluctuate, the trend over time counts, not the single measurement. A screenshot of one answer is not a result.
What can be steered is the supply. With the Google-Extended token you decide whether crawled content may be used to train future Gemini models; it has no effect on inclusion in Google Search and is not used as a ranking signal. Which is exactly why this is a business decision, not a technical one: visibility in answers and protection of your own content cannot both be had in full.
Underneath, the craft remains. Google lists several crawlers with their own tokens, among them Googlebot for smartphone and desktop, Googlebot-Image, Googlebot-Video, Googlebot-News, Google StoreBot, Google-InspectionTool and GoogleOther. Lock one of them out by accident and you lose exactly that channel. And a page that is blocked in robots.txt while also carrying a noindex stays in the index, because Google cannot read the noindex.
The questions we work through with you
For which questions are you already the obvious source today?
How you can tellThere is a wish list of keywords, but nobody has looked at the queries the site already appears for. Sorting goes by search volume, and at the top sit short, generic terms with a low click-through rate because the intent behind them is a different one.
What we doWe start with what exists: the Search Console performance report shows real queries per page with clicks, impressions, click-through rate and average position. We look for the spots with many impressions and few clicks, because that is demand you already reach but do not convert. This gets cross-checked against the search terms in your ad account that led to conversions: documented, not estimated.
Are you cited in AI answers, and for which questions?
How you can tellWhat gets measured is whether your own brand name appears. More interesting is who gets cited for the questions where nobody thinks of you yet. A second sign: the pages carry no author, no source and no date, although Google names exactly these as part of helpful content.
What we doWe build a fixed question catalog in several phrasings and run it on a fixed rhythm, recording which sources are cited. What gets evaluated is the trend. In parallel we make content verifiable: author, sources and date belong visibly on the page, because Google names trustworthiness as the most important element of E-E-A-T.
Can Google even process your content completely?
How you can tellPages are listed as not indexed in Search Console although they look reachable. Or the source code contains almost nothing and the content only appears in the browser; in the rendered version exactly this content is missing. After a relaunch, redirect chains join in, with old addresses pointing to the new one across several hops.
What we doWe check by pattern, not by page: the inventory is collected completely and grouped by template and page type, because the same error in one template costs ten thousand pages. Then we look at which crawler tokens actually access the site instead of only reading the rules file. Blocks, redirects, canonicals and noindex are read together; one signal has to win, and deliberately so.
Where between impression and click is your biggest loss?
How you can tellThere is a ranking report, but nobody knows the impressions behind it. A position without demand behind it is not an improvement. Pages with many impressions and strikingly few clicks are not on the list, although the fastest win often sits right there.
What we doWe first secure the baseline from the performance report per page and per query before anything is changed. Then the order applies: reachability before content before presentation. Improving content that is not even indexed achieves nothing. Hundreds of individual alerts become a few root causes with effort and expected impact next to them.
Are your country pages competing against each other?
How you can tellThe performance report shows several of your own pages for the same query. Common technical causes: one-sided annotations where the German page points to the English one and not the other way round, relative addresses in the markup, or a country code without a language.
What we doGoogle names HTML tags in the head, HTTP headers and sitemap entries as equally valid routes. We pick one and implement it consistently instead of running several in parallel. Reciprocity is checked by machine: every version has to list itself and all others with fully qualified addresses. More important than the markup remains the question of which markets deserve their own content, because a translated page without its own demand analysis ends up competing with the original.
The expertise we combine for this
- SEO & GEOMakes sure Google finds, processes and serves your pages, and enforces the order reachability, content, presentation.
- LLMO & GEO AgencyBuilds the question catalog and the repeated measurement that turns a snapshot into a trend.
- SEA / Paid SearchDelivers the search terms with conversions that show which intent actually pays.
- Creative & ContentWrites content that answers one question unambiguously, with verifiable statements and recognizable authorship.
And the technology to match
The same questions, asked on a fixed rhythm, in many phrasings, with the cited sources recorded: that is legwork no human sustains manually. With us it runs automated, and the evaluation stays with the team.
Evidence from client work
Frequently asked questions
How do you measure visibility in language models?
By asking the same questions regularly and in many phrasings and recording which sources are cited. Because answers vary, the trend over time counts rather than the single measurement.
What does Google-Extended do?
It controls whether content crawled by Google may be used to train future Gemini models, and covers the Gemini apps, the Vertex AI API for Gemini and grounding. It has no effect on inclusion in Google Search and is not used as a ranking signal.
What helps to get cited in AI answers?
Content that answers one question unambiguously, with verifiable statements and recognizable authorship. Google describes for helpful content that the who, how and why should be clear, and names trustworthiness as the most important element of E-E-A-T.
Is high search volume a good criterion?
Not on its own. What decides is the intent behind the term and how close it sits to your business. A small term with clear buying intent often carries more than a big one without.
Why does a blocked page stay in the index?
Because a block in robots.txt prevents reading. Google then cannot see a noindex on that page. Whoever wants a page out of the index has to leave it readable.
Is a translation enough for a new market?
Technically yes, commercially rarely. Every market searches differently, and a translated page without its own demand analysis ends up competing with the original. Fewer markets with real depth beat many with a translated surface.
Further reading
The special topics cover individual points in detail.
- Are you cited in AI answers, and if so, how often?
- Your rankings have not moved in months
- Your content is good, but Google does not show it
- You rank for terms nobody buys
- Your country pages compete against each other instead of against the competition
- Is AI Max worth it for your search campaigns?
From our Insider articles: