The 5-Step SEO Strategy for AI Search (2026)

I rank number one in Australia for AI SEO and Google still does not recommend me. It recommends the agency I work for. That one sentence explains more about SEO strategy for AI search than most GEO decks, and we get to it in step 4.

The strategy on this page is the one I run on this site and across 300+ client sites at StudioHawk. Every number in it comes from my own dashboards or a named study, and where a step is still an open experiment on my own site, I say so.

The five steps run in order because each one is a precondition for the next. A page nobody can crawl never gets extracted. A page nobody extracts never gets cited. A brand nobody cites never gets recommended.

Step 1: Can AI systems actually read your site?

AI systems read your site when three layers agree: your robots.txt, your CDN's bot rules, and your rendering. On most sites I audit, at least one of them disagrees with the other two, and nobody in marketing knows.

The failure is rarely deliberate. A security team turned on a blanket bot rule during a scraping scare. A robots.txt allows GPTBot but the firewall challenges it. A React component holds half the page's content and the crawler never runs the JavaScript. You look fine in Google, so everyone assumes the AI layer is fine too.

My own Cloudflare data makes the stakes concrete. Of the AI traffic hitting this site, the user-triggered agents matter most: when someone asks ChatGPT a question and it fetches pages live, that visit arrives as ChatGPT-User, and in my July audit it was 31.2% of operator traffic. Block that agent and you are invisible at the exact moment a human is asking about you. Meanwhile GPTBot, the training crawler everyone argues about, barely visited at all.

What to do, in order:

• Reconcile robots.txt against your CDN. Read both. If Cloudflare or your WAF has an AI-bot rule your robots.txt does not mention, one of them is lying to you. Know the three bot classes: training crawlers, search indexers, and live user-triggered agents. You can reasonably block the first and allow the last two.
• Declare your position machine-readably. This site runs content signals in robots.txt (search yes, ai-input yes, ai-train yes). Yours can differ. Pick a position on purpose.
• Check what renders without JavaScript. Fetch your money pages with curl and read what comes back. If the answer is missing, so are you.
• Ping the index that feeds the answers. Bing powers ChatGPT Search and Copilot, and Bing consumes IndexNow. I wired IndexNow into this site's deploy pipeline in August, so every content change pings Bing the minute it ships. Google ignores IndexNow; that is fine, it is not the target.

Once the bots can read you, the question becomes what they can lift off the page: extraction.

Step 2: How do you structure content so AI can extract it?

You structure content for AI extraction by making every section a self-contained passage: a question as the heading, the answer in the first sentence, and one number backing it, all inside roughly 50 words that survive being lifted out of context.

AI answers are assembled from passages, not pages. If your answer starts three paragraphs after the heading, behind a warm-up and a definition, the engine either lifts someone else's passage or lifts your warm-up. Neither gets you cited.

The format rules I apply to every page on this site:

• Phrase H2s as the questions people actually ask. Engines convert headings to questions anyway; do it for them.
• Answer in the first sentence, restating the question's own terms. Context and caveats come after.
• Bold the answer, not the keyword. A bolded search term helps nobody; a bolded verdict is what the skimmer and the model both take away.
• Keep claims specific. "A 60% lift in organic sessions" is citable evidence. "Great results" is filler.

Two honest caveats from my own data. First, this structure is under live test on this site right now: eight definitional pages rewritten this way against a matched control, readout on 8 September. I publish the result either way. Second, my July correlation re-baseline across the portfolio found that the breadth of genuine topic vocabulary on a page is the strongest on-page factor I measure (r=-0.29, up from -0.22 in April). Extraction formatting gets you lifted; vocabulary depth gets you ranked. You want both on the same page.

Structure decides whether an engine can quote you. The next question is whether it does, which is what citations measure.

Step 3: How do you win AI citations?

You win AI citations by treating them as market share per query family: find the query families where AI already grounds on your pages, defend the ones you lead, and close the gaps where you appear but rarely.

Most teams have never seen this data for their own site. Bing Webmaster Tools reports which queries ground AI answers on your pages and how often. When I pulled mine, it showed 17,211 citations across 313 grounding queries, and the distribution was nothing like my traffic report. My WooCommerce SEO guide holds a 66% citation share for its query family. The enterprise SEO tools cluster peaks at 85%. Neither is anywhere near my top pages for clicks.

That mismatch is the strategy. Citation share responds to different work than rankings do:

• Defend your leaders. A page with a 60%+ citation share is an asset. Keep it fresh, keep its tables extractable, and do not let a redesign bury the answer.
• Close the near-miss gaps. A family where you hold 5-15% share with real volume is the AI equivalent of striking distance. One stronger answer-first passage often moves it.
• Cover the fan-out. Engines answer a question by running background subqueries around it. A page that answers the head question plus its follow-ups, each under its own heading, gets picked for more of those subqueries. My FAQ blocks with anchor links exist for exactly this.
• Feed the right index. The citations above come through Bing's index. That is where ChatGPT Search and Copilot shop. Bing Webmaster Tools takes ten minutes to set up and most of your competitors have not bothered.

Citations prove the machine quotes you. A recommendation is a bigger prize: the machine naming you as the answer. That is where I have the scar tissue.

Step 4: How do you get AI to recommend you instead of just cite you?

AI recommends whoever the evidence agrees on. Engines cross-check their existing knowledge, the ranking results, and what multiple trusted third parties say, and they recommend the entity that shows up consistently in all of them. I learned this the expensive way.

Here is my own scoreboard. StudioHawk, the agency where I am Chief of Staff, ranks number one in Australia for AI SEO. I built a large part of that. But when I mined the consultant SERPs in July, my personal site was absent from both: not in the top 8 in Australia, not in the top 100 in the US. My money page sat at position 60-90 with every on-page box ticked. Google had resolved the "AI SEO" entity to the agency, and my years of work fed its authority, not mine.

The people who did break into those SERPs told me what actually works. The US number one is a solo consultant whose site has a fraction of the link authority of the brands below him, and the AI Overview cites him by name. Three other individuals rank alongside him the same way. None of them out-linked anyone. They out-mentioned everyone: consistent bios, third-party listicles, reviews, podcasts, video. Entity signals, not backlinks.

The numbers agree with the anecdote. Ahrefs' 75,000-brand study found unlinked brand mentions correlate with AI visibility at 0.664 while backlinks sit at 0.218. Three times the weight, no link required. And before anyone declares links dead: my July re-baseline found referring-domain correlation with traditional rankings doubled in six months (r=-0.16 to -0.26). The honest read is that the two channels diverged. Google rankings lean on links; AI recommendations lean on mentions. Fund both, but stop spending mention-money on links.

What I changed after the diagnosis, in the order I did it:

• Wrote the one-sentence brand fact and repeated it everywhere. Who I am, what I do, for whom, and the differentiator, identical on my homepage, my about page, my author bios, my LinkedIn, and every profile I control.
• Locked the entity claim into the money page. One H1 carrying the exact phrases I want owned, after an experiment taught me that breaking the exact phrase on a page that held position 1 was a self-inflicted wound.
• Started the third-party campaign. Pitches to the listicles that dominate the informational SERP, real testimonials queued for review markup, and my videos retitled toward the exact query family so the video carousel, a surface rankings cannot buy, carries my name.

None of that is a hack. It is reputation work with a measurement layer, which is why the fifth step exists: if you cannot see all three channels, you cannot tell whether any of this is working.

Step 5: How do you measure an AI search strategy?

You measure an AI search strategy on three separate scoreboards: Google clicks, Bing clicks, and AI citations, because the same page now posts different results on each and any single one of them lies to you.

The page that taught me this is my most-cited article. Its Google clicks fell from 1,776 a month at peak to 88 as AI Overviews absorbed the query. Same page, same period, Bing delivered roughly 20 times the Google clicks. And in the AI channel it kept getting cited and referred. One scoreboard says the page died. Two say it changed jobs.

The working setup, all of it free or nearly:

• Google: Search Console, watched for position and the click-through decay that AI Overviews cause on informational queries. Expect impressions to inflate while clicks thin out; treat impression growth alone as noise.
• Bing: Webmaster Tools for clicks and the AI citation report from step 3. This is the scoreboard most sites have never opened, and it is the one closest to the AI answers.
• Bot telemetry: Cloudflare (or your CDN) showing which AI agents fetch which pages. The pattern that matters is the lifecycle flip: a page fetched by training crawlers is being ingested; a page fetched by assistant agents is being used in live answers. Watching a page flip from crawler-heavy to assistant-heavy is watching it get hired.
• Referrals: utm_source=chatgpt.com and its cousins in your analytics, the only place AI traffic self-identifies.

Cadence matters more than tooling. I re-baseline correlations twice a year, review citation share monthly, and check the bot mix weekly. When my AI-bot traffic halved in July, the weekly check caught it, the per-bot breakdown showed it was operator-side behaviour rather than anything I broke, and I avoided burning a fortnight fixing a problem I did not have.

Which SEO levers matter at each step?

The levers at every step are traditional SEO skills pointed at new targets. The table is the whole strategy in one view.

FAQs About SEO Strategy for AI Search

What are the best AI SEO strategies in 2026?

The best AI SEO strategies in 2026 are the five on this page: verified bot access, answer-first content structure, citation-share tracking, entity and mention building, and three-channel measurement. Ranked by leverage per hour, entity work beats everything else for commercial queries, and bot access is the cheapest fix with the highest downside if you skip it.

Is GEO actually different from SEO?

GEO reuses SEO's levers but changes the targets, so it is a variation, not a separate discipline. Crawl control, content structure, digital PR and measurement all carry over; what changes is what you point them at: live agents instead of only Googlebot, passages instead of pages, mentions instead of links, and citations instead of only clicks. Anyone selling GEO as a clean break is selling a rebrand.

Do you need an llms.txt file for AI search?

No, llms.txt is not required for AI visibility, and no major engine has confirmed consuming it. I run one on this site anyway because it costs nothing to maintain and doubles as a curated index. Treat it as a cheap maybe, never as the strategy.

How long does an AI search strategy take to work?

Steps 1 and 2 move fastest: access fixes and structure changes have shown up in my citation data within weeks. Citation share (step 3) builds over one to three months. Entity and recommendation work (step 4) is the slow lane, months of consistent mentions before engines re-resolve who you are. Measurement (step 5) starts paying on day one, because it stops you optimising the wrong channel.

Can small sites compete in AI search?

Yes, small sites compete well in AI search because passage-level extraction and entity consensus care less about domain size than classic rankings do. The solo consultant outranking big brands in the US SERP I mined is the proof: fewer links, clearer entity, better passages. My own citation leaders are individual guides, not hub pages.

If you want the commercial version of this, where the strategy is pointed at your site and I do the pointing, that is what my AI SEO consulting covers. If you want to see the measurement layer first, the AI search ROI breakdown shows the three-channel model with the raw numbers, and the AI search hub holds the rest of the cluster.