A vetted learning path for owner-operators running several local brands (built with short-term-rental cohosting in mind). We read the primary sources, put every claim through a hostile review, threw out the hype, and kept the 2–4 resources per step that are actually worth your time.
Three names for one job, and each one has its own scoreboard.
Google's own 2026 guide says optimizing for its AI features is SEO: no special files, no special schema, no "chunking". Source
Share of cited URLs that showed up in only one of ChatGPT, Perplexity and AI Overviews for the same prompt (3.7M citations). So one "AI visibility score" doesn't tell you much. Kevin Indig / Omnia data
Share of local AI-search citations that came from Google Business Profile, across 1.9M citations. Business websites made up 93% of the unique cited domains. BrightLocal, Aug–Sep 2026
All three use the same 7 stages below. They differ in which stages you do and how deep you go.
~4 weeks · ~2 hrs/week. For when you mainly need more owners and guests to find you, not a new career.
~10 weeks · 3–4 hrs/week. All 7 stages in order, including the research (GEO paper, Indig, Ahrefs schema study, SparkToro) and the multi-brand playbook. Finish with Stage 7 and run the monthly review on every brand.
Do the Fast path first, because you can't automate a process you've never done by hand. Then wire up:
See Automate this for vetted tools and prices.
Each stage has a goal, the best 2–4 resources, and one exercise to do on your own brand. ★ marks the single best link.
| Resource | Why it's worth it | Time |
|---|---|---|
| ★ Google SEO Starter Guide | The primary source, short, with the myths removed. | 1 hr |
| How Google Search Works | The three stages in plain English. | 30 min |
| Google spam policies | Read the doorway and scaled content abuse sections. Multi-brand operators trip these, and the August 2026 spam update hit scaled AI content hard. | 45 min |
| Moz Beginner's Guide (optional) | Good for vocabulary. It's secondary, so Google's docs win wherever they disagree. | 3 hrs |
| Resource | Why | Time |
|---|---|---|
| ★ Ahrefs SEO Course for Beginners | A free, tight video course covering keywords, on-page, links and technical. | ~2 hrs |
| Ahrefs keyword research guide | Intent and "traffic potential" vs. raw search volume. | 1 hr |
| Semrush Academy | Learn the tool you already pay for (free courses). | 2–3 hrs |
| Resource | Why | Time |
|---|---|---|
| ★ Google: improve your local ranking | Google's own word on relevance, distance and prominence. | 20 min |
| Business Profile guidelines | What gets profiles suspended: keyword-stuffed names, virtual offices, duplicates. | 30 min |
| BrightLocal: 200k local AI searches | New (Sep 2026). In this data, Maps rank didn't carry straight over into AI visibility, your own site was a major source, and the same prompt gave different results on repeat runs. | 20 min |
| Whitespark Local Ranking Factors | A useful expert survey. Treat it as opinion, not measurement. | 1 hr |
| Resource | Why | Time |
|---|---|---|
| ★ Google: intro to structured data | What Google actually uses schema for (rich results). | 30 min |
| Organization + LocalBusiness markup | The two types every brand homepage needs, with sameAs linking to that brand's own profiles. | 45 min |
| Ahrefs: 1,885 pages added schema; AI citations barely moved | Reality check (May 2026). A controlled study found no lift on pages that were already cited. Its effect on brand-new pages is unknown. | 15 min |
Note: Google's vacation-rental rich result is limited to eligible partners. Test markup with the Rich Results Test and the Schema Markup Validator.
| Resource | Why | Time |
|---|---|---|
| ★ Google: optimizing for generative AI features | The official 2026 guide. It covers "non-commodity content", query fan-out, and a myth list (llms.txt, chunking, fake mentions). | 30 min |
| Google: helpful, people-first content | Self-assessment questions to run every page through. | 30 min |
| Ahrefs AEO Course | A free, practical course on how AI answers pick sources and how to measure it. | ~2 hrs |
| Resource | Why | Time |
|---|---|---|
| ★ GEO paper (Aggarwal et al., 2023 / KDD 2024) | Coined "GEO". In a lab setup, adding statistics, quotations and cited sources lifted visibility by up to ~40%, and keyword stuffing didn't help. Engines have changed since, so read it as directional. | 1 hr |
| Kevin Indig: The Consensus Gap | New. Engines mostly cite different URLs, so track each engine separately. | 20 min |
| SparkToro: AIs are highly inconsistent | Why one-off checks are noise. Only visibility % across many repeated prompts means anything. | 20 min |
| OpenAI crawlers · Publisher FAQ · Perplexity crawlers | Allow OAI-SearchBot to appear in ChatGPT search (GPTBot, the training crawler, is a separate decision). ChatGPT referrals show up with utm_source=chatgpt.com. | 30 min |
| iPullRank AI Search Manual (Quick Start) | A free, practitioner-grade explainer of how each engine retrieves and cites. | 45 min |
| Resource | Why | Time |
|---|---|---|
| ★ Search Console: Generative AI performance report | Free first-party data on AI Overviews/AI Mode impressions by page, country and device. No queries, no clicks, no API, so you export the CSV by hand. | 20 min |
| Bing Webmaster Tools: AI Performance | Free Copilot citation data: cited pages, grounding queries and citation share. | 20 min |
| Semrush AI visibility features | Shows what your plan covers. AI Overview tracking is on all plans. ChatGPT/AI Mode prompt tracking needs Semrush One or the AI Visibility Toolkit. | 20 min |
| learningseo.io (Aleyda Solis) | A free, maintained roadmap for once you're past this path. | ongoing |
Every key claim was put through a hostile review. Verified means a primary source or a controlled/large dataset backs it. Contested means credible people disagree or the evidence is narrow. Hype means unsupported, contradicted or risky.
This is Google's official position. What would overturn it: Google documenting an AI-specific signal.
Rolled out to all sites Aug 31, 2026. Plan on exporting it by hand.
SparkToro found AIs almost never return the same brand list twice. BrightLocal found local recommendations shift between runs, though businesses reappear 50%+ of the time.
*Large dataset (3.7M citations), but from one vendor (Omnia), skewed toward Europe, and not peer reviewed. What would overturn it: an independent replication showing high overlap.
Ahrefs' controlled study (1,885 pages vs. 4,000 controls) found no lift for already-cited pages, and a small decline in AI Overviews. A separate test found live AI fetches ignored JSON-LD. Unknown: whether it helps pages that have never been cited. Change made: schema is now framed as hygiene and rich-results eligibility, not a GEO lever.
True in the GEO paper's lab benchmark (2023). There's no controlled replication on today's production engines. Keep the direction (evidence beats keywords) and drop the exact numbers.
Rand Fishkin argues the model often already "knows" the brands before it searches (citations are the receipt, not the decision). Lily Ray and others show ChatGPT's retrieved pages strongly predict which brands it names. Practical upshot: track mentions and citations separately.
BrightLocal (200k prompts): business sites were 93% of unique cited domains and 42% of citations, and GBP alone was 28.5%. A study built on Ahrefs Brand Radar data found directories took 36% and business sites only 6.7%. The method decides the answer, so cover both your own site and your directory profiles.
This is a community heuristic. Google says no chunking is needed. Clear, answer-first writing helps readers either way, so do it and don't count words.
Plausible and cheap to fix, but there's no controlled study. Keep it as a cheap fix, not a promise.
Early anecdotes after the Gemini 3.8 Flash rollout (Sep 2026) say AI Mode is less likely to link to sources. That's not measured yet. Watch your GSC Generative AI report rather than assuming.
Google says its Search ignores it. Other engines have published no evidence that they use it. It's harmless, but it won't move anything.
Engines barely overlap and answers vary from run to run. Use per-engine presence across repeated prompts instead. Free "AEO/GEO checker" scores are marketing.
Widely promoted on X. That's the scaled content abuse Google's August 2026 spam update targeted.
Research shows AI search can be manipulated this way. Google explicitly says inauthentic mentions aren't helpful, and its spam systems sit under AI features. High risk, short shelf life.
Spam that violates the Business Profile guidelines and risks suspension.
Agents still need data from somewhere: GSC, Semrush, Ahrefs or DataForSEO. Use agents to operate tools, not as a replacement for the data.
Tools surfaced from X discussion (Apr–Oct 2026), each checked on its own site. Prices are as shown on Oct 8, 2026, in USD. Check again before buying.
Sends your full GSC performance data to BigQuery every day, past the UI's row caps and its 16-month history limit. It only captures data from the day you turn it on, so start now.
Lets an AI agent query Search Console (queries, pages, URL inspection) across properties. ~1.9k GitHub stars and actively maintained. Alternative: search-console-mcp (also covers Bing and GA4).
SEO toolkit: $139/mo ($117.33/mo billed annually), including Position Tracking with AI Overview tracking. Semrush One Starter, which adds ChatGPT/AI Mode/Gemini prompt tracking: $199/mo ($165.17/mo annually). Pro+ $299 / Advanced $549 for more sites.
Tracks how local businesses appear in ChatGPT, AI Mode and AI Overviews, localized by coordinates, plus local rank grids and citation audits. Track plan: $41/mo for 1 location ($31/mo annually), $62/mo for 2–5 locations. Free 14-day trial.
Daily prompt tracking across 3 chosen models (ChatGPT, AI Mode, AI Overviews, Copilot, Gemini…). Shows source/citation share and gap analysis. Starter $95/mo (50 prompts, 1 project), Pro $245 (150 prompts, 2 projects), Advanced $495.
Geo-grid rank tracking for Maps/local pack, run on credits. Starter $24.99/mo or $299.88/yr. Higher tiers up to $199.99/mo.
First-party Copilot/Bing AI citation counts, grounding queries, citation share and period comparisons (preview).
No paid tool needed. Have an agent draft JSON-LD from each brand's fact sheet, then validate with the Rich Results Test and Schema Markup Validator. For briefs: export fan-out queries from prompt tracking and cluster them into topics rather than chasing each one (Lily Ray's workflow).
Dropped after checking: free "AEO/GEO score" checkers, AI-autoblogging tools, and "replace your SEO suite" agent templates (see the Evidence board).
How to run several brands without them competing with each other or tripping spam rules.
/park-city/). Don't spin up a domain per town.Organization/LocalBusiness JSON-LD per brand, each with its own @id and sameAs. Show real relationships with parentOrganization.Common advice that failed the hostile review.
About 3–4 hrs/week while learning, then a lighter maintenance rhythm.
| When | Learn | Apply |
|---|---|---|
| Mon · 30m | Main resource for the current stage | Pick one brand + one task |
| Tue–Wed · 60m | Second resource | Do the stage exercise |
| Thu · 45m | – | Ship it: publish, fix or mark up |
| Fri · 30m | Skim the Search Central blog | Update the fact sheet + keyword map |