{"id":111473,"date":"2026-10-15T16:05:23","date_gmt":"2026-10-15T14:05:23","guid":{"rendered":"https:\/\/ga.agency\/?post_type=blog&p=111473"},"modified":"2026-10-15T16:09:22","modified_gmt":"2026-10-15T14:09:22","slug":"5-lessons-searchnorwichxl-2026","status":"publish","type":"blog","link":"https:\/\/ga.agency\/en\/blog\/5-lessons-searchnorwichxl-2026\/","title":{"rendered":"Five things we learnt at SearchNorwichXL 2026"},"content":{"rendered":"\n
<\/div>\n\n\n\n

On 24th September 2026, our Tech Lead Francesca spent the day at SearchNorwichXL<\/strong> – held at Norwich’s Maddermarket Theatre, a Grade II listed playhouse that opened in 1921 as the first permanent recreation of an Elizabethan theatre.<\/p>\n\n\n\n

Nine sessions covered everything from People Also Ask data to the long-term health of the web. We have pulled out the five lessons that matter most for the brands we work with and the common thread was clear: the brands best placed for AI-driven discovery are the ones whose foundations are already in order.<\/strong><\/p>\n\n\n\n

<\/div>\n\n\n\n
\n
\"\"<\/figure>
\n

1. Search data is a snapshot, not a fact<\/strong><\/h3>\n<\/div><\/div>\n<\/div><\/div>\n\n\n\n
<\/div>\n\n\n\n

Arnout Hellemans opened with a study of Google’s People Also Ask (PAA)<\/strong> boxes: 340 question trees in seven languages, covering Heineken’s 11 most important markets and the top ten cities in each. The data showed how differently people approach the same brand. In Nigeria, whether Heineken counts as beer or alcohol was a top-level question. In the US, a question about the worst beer appeared in all ten cities on every collection date.<\/p>\n\n\n\n

The more useful finding was how unstable that data is. Google appears to rotate a handful of questions from a larger pool, so results shift between pulls. A link between Peroni and Heineken surfaced in five Italian cities on the first date, none on the second and seven on the third. Roughly one pull in twelve came back in English without warning and of eight single-location claims tested, none survived a second pull.<\/p>\n\n\n\n

\u27a1\ufe0f SERP data can signal sentiment, but a single snapshot should never shape strategy or budget.<\/strong> Freshness requirements, language settings and cross-checks across locations and dates need defining first. In multi-market work, this is also where native-language review earns its place. A result returned in the wrong language is easy to miss when nobody on the team reads that market’s language.<\/p>\n\n\n\n

<\/div>\n\n\n\n
\n
\n

2. Being seen is not the same as being remembered<\/strong><\/strong><\/h3>\n<\/div>
\"\"<\/figure><\/div>\n<\/div><\/div>\n\n\n\n

Two sessions reached this point from different directions. Beng\u00fc Sar\u0131ca Din\u00e7er traced how a single need now moves across TikTok, Reddit, ChatGPT and other assistants before someone turns to Google or types a brand’s URL. Clearer content beats more content and single-touch attribution in GA4 misreads this kind of journey.<\/p>\n\n\n\n

Giulia Panozzo approached it through consumer neuroscience, including research measuring skin conductance. Much of a click or purchase is driven by emotion and prior association rather than by the page itself. Leaning too heavily on AI-produced copy risks flat brand experiences and Gen Z audiences in particular respond to authenticity and shared values.<\/p>\n\n\n\n

\u27a1\ufe0f Brand recognition now belongs alongside rankings as something to measure.<\/strong> That requires attribution that looks beyond the last click and messaging that holds together across every platform a customer passes through.<\/p>\n\n\n\n

<\/div>\n\n\n\n
\n
\"\"<\/figure>
\n

3. AI visibility starts with what models already believe<\/strong><\/strong><\/h3>\n<\/div><\/div>\n<\/div><\/div>\n\n\n\n
<\/div>\n\n\n\n

Dixon Jones argued that the first job is an entity gap analysis<\/strong>: comparing how language models understand a brand with what the brand says about itself.<\/p>\n\n\n\n

His most practical tip was to switch off live web lookups when testing models. That way the test shows what a model learned in training and not what it retrieved that day. Hallucinations often trace back to outdated or third-party pages, so correcting them means fixing those sources, including the listicles and external pages models cite. Share of voice is more meaningful when measured as topic coverage against every competitor that appears alongside the brand.<\/p>\n\n\n\n

Myriam Jessier extended this beyond text. Video, images and audio increasingly answer questions about a product’s size, packaging or use. Machines read these assets through technical signals such as object detection and how closely an asset matches its surrounding context.<\/p>\n\n\n\n

\u27a1\ufe0f AI visibility is a diagnostic problem.<\/strong> Producing more content without the audit risks reinforcing what models already get wrong.<\/p>\n\n\n\n

<\/div>\n\n\n\n
\n
\n

4. Indexing is earned continuously and lost quietly<\/strong><\/strong><\/strong><\/h3>\n<\/div>
\"\"<\/figure><\/div>\n<\/div><\/div>\n\n\n\n
<\/div>\n\n\n\n

Adam Gent drew on Google antitrust trial material, Google patents and data from millions of URLs checked through the URL Inspection API. His argument was that indexing is an ongoing quality process<\/strong>, not a question of whether a page can be crawled. Google keeps removing pages against shifting quality thresholds and a lack of user engagement is often the main reason a page drops out.<\/p>\n\n\n\n

The data was striking. A study of 1.4 million pages across 18 websites found that a page not crawled in 130 days had a 99% chance of not being indexed. A page crawled in the last 30 days had a 97% chance of being indexed. After around 190 days without a crawl, pages tended to move to URL is unknown to Google<\/em>.<\/p>\n\n\n\n

\u27a1\ufe0f For large sites, days since last crawl is an early-warning metric.<\/strong> Tracking it across important templates catches decline before it shows up in traffic.<\/p>\n\n\n\n

<\/div>\n\n\n\n
\n
\"\"<\/figure>
\n

5. Technical SEO has to sit inside the release process<\/strong><\/strong><\/strong><\/h3>\n<\/div><\/div>\n<\/div><\/div>\n\n\n\n
<\/div>\n\n\n\n

Small code releases routinely break canonicals, meta tags and indexing directives. Developers and SEO teams work in separate loops, so the damage is usually spotted only once it is live.<\/p>\n\n\n\n

Estela Franco, Senior Technical Architect in Web Performance at Shopify, presented an approach she calls SEO as Code<\/em><\/strong>.<\/em> Expected SEO states are written as machine-readable JSON contracts, generated from a site crawl, then checked automatically with Playwright inside GitHub Actions. Rendering problems and serious bugs are flagged on the pull request before anything merges.<\/p>\n\n\n\n

\u27a1\ufe0f SEO checks belong in the deployment pipeline.<\/strong> For sites with frequent releases, it lowers risk and lets teams ship faster.<\/p>\n\n\n\n

<\/div>\n\n\n\n

<\/span>What we are bringing back<\/strong><\/span><\/h2>\n\n\n\n
<\/div>\n\n\n\n

The day closed with Martin Splitt, of Google’s Search Relations team, speaking in a personal capacity as someone who has used and built for the web for many years. He warned against reaching for generative AI as a patch for poor user journeys, citing its unpredictable output, its running costs and the open question of accountability. The closing panel reached the same conclusion from another angle: generic AI content at scale is a dead end.<\/p>\n\n\n\n

Across the sessions, the brands positioned to gain from AI-driven discovery were not the ones producing the most content. They were the ones whose technical foundations, entity data and brand signals already told a coherent story. These ideas are now feeding into our technical roadmaps, from indexing monitoring on large sites to entity and citation audits for AI visibility.<\/p>\n\n\n\n

If any of these questions are live for your brand, get in touch with our team<\/a>.<\/p>\n

\n
\n<\/span><\/div>\n