AI SearchBootcamp

Recap: How AI search engines build their answers, the concepts that matter, and how to make sure AI crawlers can reach your site at all.

• 11:55: One run is a coin toss. AI answers are probabilistic. Run each prompt 7 to 10 times before you trust a visibility score.
• 12:18: Four metrics to act on: mention, sentiment, citation, and source. A source is used but not credited, a citation is credited.
• 13:41: Three layers, three speeds. Training data changes with each new model, user context changes instantly, live web search changes within days. New content can move the web layer fast.
• 16:07: Sentiment decides whether you get recommended. AI engines act like salespeople: if they only hear bad things, or find nothing, they steer buyers away.
• 17:02: More content is not the answer. Blasting out AI pages or self-promotional listicles gets penalised, and engines drop the self-ranked brand from the answer.
• 25:08: Query fan-out: one prompt triggers many hidden sub-searches. You do not need to rank for one query, you need to be findable across the fan-out.
• 28:26: Split branded from non-branded. Branded visibility should sit above 95%. Non-branded is the real battleground.
• 29:53: Self-report beats referral tracking. Rankscale AI sees 5x more customers say they found us via AI than AI referrals show.
• 30:42: Four ways to be invisible: a web application firewall, robots.txt, JavaScript rendering, and not being indexed in Google, Bing, or Brave. Put critical content in the HTML.
• 36:47: The passage is judged, not the page. A small brand with one well-aimed paragraph can beat a big brand. Backlinks matter far less than SEO intuition says.

Recap: How to build a tagged prompt group, pick engines and tracking frequency, and use seven metrics to find the prompt to fix first.

• 01:22: One core prompt, 12 to 48 variations, built from four variables: branded or not, funnel phase, scope or geography, and persona. Tag every variable so you can filter.
• 02:08: Keep prompts short. No filler or question words: every extra word adds noise, and engines strip them from their own sub-queries anyway.
• 04:07: European brands: track English too. AI engines often run their sub-queries in English, which favours brands with English content.
• 11:37: Where real prompts come from: Prompt Research, a Google Search Console regex filter, sales calls, support tickets, G2, Capterra, Trustpilot, and competitor FAQ pages.
• 14:53: Engines follow the buyer. Enterprise: Copilot, ChatGPT, AI Overviews, Claude. SMB: ChatGPT, Copilot, AI Overviews. Consumers: ChatGPT and AI Overviews. Add Mistral for France, Claude and Grok for developer tools, Perplexity for marketers.
• 21:48: Wait for seven runs before you read the data or start fixing.
• 23:22: Benchmarks: branded visibility above 95% (below 90% is an entity problem), sentiment above 65% means endorsed, below 55% is a warning, average position worse than 7 is a red flag.
• 26:11: Read the pattern: high mentions with low sources is a content gap, low mentions with high sources is a brand attribution gap, both low is an entity emergency.
• 32:53: The Citations tab shows why you are invisible. In the Drei example, a competitor YouTube video with only 12k views was what Google AI Overviews cited.

Recap: How to find the content AI engines cite instead of yours, rewrite pages so they get quoted, and build a grounding page.

• 01:12: Two principles to judge any tactic: token economics (content that is cheaper to process gets favoured) and game theory (serve the engine's goal of an objective answer, do not try to game it).
• 03:31: It works: the experiment site howtoaustria.com covers 74% of hundreds of tracked prompts by following these rules.
• 04:49: Three ways to find gaps: the Citations tab, competitors' top URLs in the Sources tab, and exporting full AI responses into an AI assistant to group what keeps getting cited.
• 17:00: Six gaps that stop citations: a 40 to 80 word bottom line up front (BLUF), justification attributes, freshness, evidence and data, FAQ, and the answer near the top.
• 17:56: Freshness and data pay off. Pages updated in the last 30 days earn up to 3x more citations, quantified claims up to 40% more visibility, and most citations come from the first 35% of a page.
• 20:12: Justification attributes counter market complaints. Aura Frames repeats "no subscription fees" everywhere because that is the category's top complaint.
• 21:00: Write headers as questions and put a number on every claim.
• 22:39: AI drafts, people own it: strategy, brand voice, positioning, and E-E-A-T stay human. Remove the tell-tale AI phrasing before you publish.
• 29:12: Grounding pages are factual, machine-readable pages that remove four risks: hallucinations, entity confusion, non-inclusion, and English retrieval bias.

Recap: How to make every page easy for AI engines to reach, read, and quote.

• 00:57: Four ways a page fails the machine: crawl success, semantic structure, schema markup, and chunkability.
• 02:06: Cheaper to read means more likely cited. Engines follow token economics.
• 02:34: One H1, never skip a heading level. A broken hierarchy produces broken retrieval chunks. Check it with View Source, not the browser.
• 03:39: The accessibility tree is the agent's map. Browser agents navigate by real headings, buttons, landmarks, and alt text.
• 04:18: Declare, do not make engines guess: Article, FAQPage, HowTo, Product or Offer, LocalBusiness, and Organization schema, in JSON-LD that mirrors the visible text.
• 05:44: One name per entity, everywhere. Aura Frames keeps "Aura Frame" and "Aura app" consistent across 432 pages. Check yours with a site: search.
• 12:49: Every URL follows one structure: nav around the menu, content in article, one section per topic, dates in a time element.

Recap: What other sites say about you now shapes whether AI engines recommend you. How to close entity gaps, get onto the pages AI trusts, and correct false claims.

• 01:37: Four PR gaps: entity, third-party citations, sentiment and reputation, and authority.
• 03:16: Join the sources of truth: Wikidata, LinkedIn, Facebook, X, Crunchbase, and the Google Knowledge Panel, plus your own grounding page. A grounding page can even outrank the official page on Google.
• 04:35: Hunt false information on purpose. Track prompts like "why should I not use [brand] as an expat" under one tag, then follow each negative claim in the Sentiment tab to the page behind it and ask the publisher to correct it. Hotel Sacher fixed a false "no pool" claim this way.
• 08:38: Find the pages to get on: filter the Citations tab without your brand. Every domain cited without you is an outreach target.
• 10:24: Four places to get cited: partner ecosystems (G2, Salesforce AppExchange), editorial (Forbes, Wikipedia, HARO), community (Reddit, YouTube, TikTok), and niche blogs.
• 11:44: One identical description on every profile, so engines know what your brand stands for.
• 12:43: Prioritise by citation share, and by URL over domain. The cited URL is what moves the answer. Backlinks and domain authority matter far less than in classic SEO.

Recap: How to turn Modules 1 to 5 into an operating rhythm with five cadences and one owner each.

• 00:00: GEO fails when nobody owns the cadence. Name an owner, set the rhythm, write it down.
• 01:00: Five cadences: weekly, each contributor ships one named priority. Biweekly, the GEO lead sets the next priorities. Monthly, the ROI readout. Every two months, new sub-queries. Quarterly, a full review.
• 01:00: One GEO lead, three contributors: SEO and readability, content, and PR and entity. Working solo means wearing all four hats.
• 05:19: Set up once: reachability, two primary engines, money prompts first, and a seven-day baseline with GA4 and self-attribution wired in before anyone changes a page.
• 07:19: Three priorities, in order: improve content that already wins, create new winning content, then get onto the third-party pages that matter.
• 08:20: Never delete a row from the action log, even after a quiet week. It is what justifies the budget later.
• 11:33: Flag any 10-point drop on a money prompt, and focus on citations that appear across more than one engine.
• 14:28: Check the WAF monthly. CDN providers sometimes change policies and start blocking AI crawlers.
• 17:27: Buy placements one quarter at a time, then recheck whether the page is still cited.

Recap: How to prove what AI search work is worth, report it to each audience, and build the dashboards that keep the budget coming.

• 01:44: Four attribution paths: AI referral clicks, branded search lift, direct traffic lift, and self-reported attribution.
• 03:17: Ask "how did you hear about us?" Five times as many Rankscale AI customers report finding us via AI as AI referral tracking shows.
• 04:27: One report per audience: board or client monthly, marketing lead biweekly, practitioner weekly. If everyone gets the same report, you are doing it wrong.
• 06:38: Four rules for every report: one metric that tells the story, anchored to a competitor or benchmark, the input behind it, and the next action.
• 08:00: A number needs context. 12% visibility means nothing; 12% against a peer median of 18% is a gap you can close.
• 09:40: Four dashboards: funnel health, niche dominance, engine gap, and scenario proof on long-tail prompts, where wins come fastest.
• 17:32: Report the inputs too. Mark actions done in the Recommendations tab so the board sees what moved the number.

Recap: What agencies charge for AI search work in DACH and the US, how to package it, and how to win the first meeting on the prospect's own data.

• 00:37: Six offers: snapshot or audit, recommendations, 90-day pilot, content and guest articles, mention building, and retainer. A snapshot runs €790 to €4,500 in DACH ($1,500 to $7,500 in the US).
• 06:03: Never open with a retainer. A snapshot earns a pilot, a pilot earns a retainer.
• 06:35: Price from hours, tooling, and margin, so the number is easy to defend, and credit the snapshot fee against the pilot.
• 09:37: Six signals a prospect needs you, from low non-branded visibility on their core product to being strong on the wrong engine.
• 16:54: Run the first meeting live. Ask ChatGPT the buyer's question in front of them and point at who gets named. People believe what they see.
• 18:20: Sell the gap, not GEO. They buy being visible when a prospect looks for a solution.
• 19:39: "Can you guarantee we show up?" Nobody can, answers are probabilistic. Commit to the work, the tracking, and a day-90 readout against written criteria.