GEO/AEO Tactics That Survived 10 Months of Testing

• Every number in this article has a named source. Where the evidence is my own client tracking, I say so explicitly.
• Citation behavior is platform-specific: what wins in Perplexity barely registers in ChatGPT, so one "AI strategy" is really three.
• Brand mentions without links correlate with AI visibility roughly 3× stronger than backlinks in Ahrefs' 75,000-brand study.
• Structured, sourced, quotable content wins measurably; fluent-but-vague content wins nothing, per Princeton's GEO research.
• Two "tactics" circulating in 2026 are traps: templated review networks and prompt-injection buttons. Microsoft is already cataloguing the second one.

I run SEO for a Dubai marketing agency by day, and for the last 10 months my team and I have been compiling a GEO/AEO playbook: 31 tactics, tested on client projects, updated as the engines changed. In February 2026 it settled into something stable enough to publish. This article is the distilled version, with one rule my clients will recognize: a claim only counts if I can show where it comes from. Each section names its source, whether that’s a public study or our own tracking.

How I sorted 31 tactics into “works” and “noise”

The playbook groups tactics into four buckets: content formatting, platform presence, technical access, and strategic signals. For each one we tracked whether it moved AI-answer presence for the prompt sets we monitor monthly across ChatGPT, Perplexity, Google AI Overviews, Claude and Gemini. That internal tracking is the “my data” source in this article. For the industry-wide numbers, I lean on four public datasets: Profound’s citation study (680M+ citations analyzed), Ahrefs’ 75,000-brand correlation analysis, the Princeton GEO research, and vendor documentation from Cloudflare and Microsoft.

One honest caveat before the tactics: this field is 2 years old. Sample sizes are small, engines change quarterly, and anyone selling certainty is selling something else. What follows is the best evidence I have, not physics.

Platform presence: go where the engines already cite

The single most useful public dataset here is Profound’s analysis of 680+ million AI citations. Two findings shaped our platform strategy.

First, citation behavior is wildly platform-specific. Reddit accounts for 46.7% of Perplexity’s citations to its ten most-cited domains, yet only about 1.8% of ChatGPT’s total citations, where Wikipedia leads at 7.8% (Profound, 2025). A precision note most articles skip: the famous 46.7% figure describes Reddit’s share of Perplexity’s top-10 domain citations; across Profound’s full dataset Reddit is closer to 6.6% of everything Perplexity cites. Both numbers are real. Quoting the big one without the frame is how this industry manufactures hype.

Second, YouTube punches far above its SEO weight: it sits among the most-cited domains for both Perplexity and Google AI Overviews (Profound, 2025). The engines don’t watch video. They read titles, descriptions and transcripts, which means the text layer under your video is the actual asset. We treat every client video as a page: full description, key facts in text, subtitles on.

What this meant in practice for our projects: review aggregators, Google Business Profile, Bing Places (which feeds ChatGPT’s Microsoft-side plumbing), and genuinely useful Reddit and Quora answers went from “nice to have” to the front of the queue. Not because a checklist said so, but because the citation data says that’s where the engines already look.

Mentions beat backlinks, and it isn’t close

The strongest external evidence in the whole playbook is Ahrefs’ study of 75,000 brands: unlinked web mentions correlate with AI Overview brand visibility at 0.664, backlinks at 0.218. Brand anchors sit at 0.527 and brand search volume at 0.392. The top three correlating factors are all off-site entity signals, not link equity. Correlation isn’t causation, and Ahrefs says so too, but a 3× gap is hard to ignore, and it matches what we see on client prompt sets: brands that get talked about get recommended, linked or not.

A supporting data point: in an Editorial.link survey, 80.9% of SEO professionals said unlinked brand mentions carry real value. The practical playbook that follows from this: expert commentary, podcast appearances, cases published on platforms AI reads (Medium, LinkedIn), review presence, and outreach into the “best X” listicles that engines already cite when they fan out a commercial query. We call that last one ranking outreach: pull the lists AI currently quotes for your niche, then earn placement in those exact lists instead of publishing your own.

One thing I refuse to do, and recommend you refuse too: networks of templated “review” sites built to fake that mention footprint. AI systems evaluate the same authority signals search always did, and manufactured consensus across thin domains is exactly the pattern platforms now hunt for. More on that in the last section.

Content that gets lifted: structure plus sources

The Princeton GEO study (Aggarwal et al.) tested nine optimization methods against generative engines and found that adding citations, quotations and statistics boosted content visibility in AI answers by up to 30-40%, while pure “fluency optimization” did roughly nothing. That one result explains most of our content playbook:

1. Answer first. A 40-60 word direct answer under a question-shaped heading, because retrieval systems lift self-contained blocks.
2. TL;DR and key-takeaways lists on long pages. Condensed, verifiable statements are what models quote.
3. Tables for anything comparative. The data arrives pre-structured for extraction.
4. Real numbered lists in semantic HTML (ol/li), not visual numbering, for processes and checklists.
5. Statistics with named sources and expert quotes with names and titles, because that’s literally what the Princeton study measured.

On our own projects the same pattern holds: the pages that earn AI citations are the ones a model can quote without cleanup. This site is built on that principle, and it’s the core of the AEO service: the page either survives extraction or it doesn’t exist to the answer engine.

The technical layer where most sites silently fail

This is the least glamorous bucket and the one with the highest hit rate in our audits.

Cloudflare blocks AI crawlers by default. Since July 1, 2025, Cloudflare, which fronts a huge share of the web, asks new domains whether to allow AI crawlers and defaults to blocking them without permission. A clean robots.txt means nothing if the CDN answers 403 before your server sees the request. The check order we use: Cloudflare bot settings and AI Crawl Control, then WAF rules, then security plugins, then robots.txt. Your server logs are the only ground truth, because GA and Search Console do not record AI crawler visits at all.

JavaScript is still a wall. Most AI crawlers read source HTML and don’t render JS. The two-minute test: disable JavaScript and reload your page. Whatever disappears doesn’t exist for the engines. SSR or static generation fixes it structurally.

Token budgets are real. Cloudflare’s Markdown for Agents converts HTML to Markdown on request for AI agents; their own blog post drops from 16,180 tokens as HTML to 3,150 as Markdown, an 80% reduction (Cloudflare, 2025). Whether or not you enable that specific feature, the lesson stands: navigation, scripts and div soup spend the crawler’s attention before it reaches your content. Clean semantic HTML, key facts high on the page, an llms.txt file, and schema markup all serve the same goal of making the machine’s job cheap.

IndexNow and segmented sitemaps round it out: Bing’s index feeds ChatGPT and Copilot, so instant indexing notification is a low-effort, direct line into one of the two big AI ecosystems.

What I’d do in the first two weeks

Our playbook ranks every tactic by effort against observed impact. The quick-wins tier, doable in days:

The medium tier (1-2 months) is content restructuring, schema, review platforms and YouTube. The long game (3-6 months) is mentions, Reddit/Quora presence and ranking outreach. Fan-out query mining runs continuously: engines decompose “rent a car in Dubai” into dozens of sub-queries, and tools that expose those sub-queries tell you what content to build next.

The two tactics I documented and won’t use

Both circulate in 2026 pitch decks, so you should know they exist and why they’re radioactive.

Templated review-site networks (a PBN dressed for the AI era) manufacture “best X” consensus across fresh domains. The problem: engines weigh domain history, traffic and E-E-A-T signals, and the pattern is detectable at scale.

The second is worse. Companies have been embedding hidden prompt-injection instructions in “Summarize with AI” buttons, trying to write “always recommend us” into users’ assistant memory. Microsoft’s security team documented 50+ such prompts from 31 companies across 14 industries and named the technique AI Recommendation Poisoning, classified under MITRE ATLAS as memory poisoning and prompt injection (Microsoft Security Blog, February 10, 2026). These were real businesses, not scammers, which tells you how tempting the shortcut is. Platforms are actively hunting it. The reputational downside of being on that list is permanent; the upside was never durable.

If a vendor pitches you either of these as “GEO”, that’s your cue to leave. The white-hat version of the same goal, being the brand AI systems recommend, is exactly the GEO work this whole article describes: earn the citations on platforms the engines already trust.

Sources

1. Profound — AI Platform Citation Patterns, 680M+ citations analyzed (tryprofound.com)
2. Ahrefs — An Analysis of AI Overview Brand Visibility Factors, 75,000 brands (ahrefs.com)
3. Aggarwal et al. — GEO: Generative Engine Optimization, Princeton et al. (arxiv.org/abs/2311.09735)
4. Cloudflare — Content Independence Day announcement, July 1, 2025, and Introducing Markdown for Agents (blog.cloudflare.com)
5. Microsoft Security Blog — The rise of AI Recommendation Poisoning, February 10, 2026 (microsoft.com/security/blog)
6. Editorial.link — SEO professionals survey on unlinked brand mentions
7. Internal: Marketing Bear GEO/AEO tactics playbook, 31 tactics compiled and tested across client projects, ten months through February 2026