The AEO Sprint

Baseline and access check

• I run your agreed buyer prompts across all eight LLMs and record the starting position.
• I pull your branded search numbers and AI referral traffic, and add a free-text "How did you hear about us?" field to your enquiry form.
• I check that AI crawlers can reach your site through robots.txt, your firewall and your CDN, and that your case studies and results sit in plain HTML, not in PDFs.
• We agree how articles get published: direct CMS access, or a publish-by date for each article.
• I also ask whether you already work with a PR agency.

A recorded start position makes the end result checkable.

The access check matters because the main AI crawlers read a page's raw HTML and do not run JavaScript. Vercel and MERJ found this across more than 500 million real crawler fetches.

OpenAI and Anthropic each run separate crawlers for search and for training, and blocking the wrong one removes you from their search answers.

Do AI crawlers render JavaScript? has the detail, and Is your website blocking the AI bots that should cite you? covers accidental blocking.

Strategy & discovery

• A discovery session on what your firm does, who it is for, what you want to be known for and the prompts you want to appear for.
• I sort those prompts into two groups: prompts where the AI searches the live web, and prompts it answers from memory.
• We leave with the prompts prioritised and three or four authority pillars drafted.

The two groups need different work. In Searchable's study of 95,000 Claude search records, "best CRM 2026" triggered a live search 9 times in 10, and "how to pick a CRM" triggered one 1 time in 34.

Articles on your own site can win the first kind. The second kind moves through sustained coverage on other sites, and I tell you which is which before I write anything. How to get cited by Claude covers the study.

Strategy confirmation

• I confirm the target phrases and propose article titles.
• You have three business days to approve, and no response counts as approval.
• A requested change moves the start date back and leaves the writing schedule alone.
• While you review, I check that your service, about and case-study pages state the same pillars in plain HTML, and I write the copy for any quick fixes.

I fix the pillars before the first article because changing them partway through resets the signal.

The page check follows Semrush's 2025 data, which puts business and service pages at 50% of ChatGPT's cited sources, against 8.3% for blogs. If your service pages say something different from your articles, a model sees two positions.

Production

• I publish six articles, one a week.
• Each one opens with its answer, carries specific numbers, names its sources, shows its questions and answers as visible text and uses the exact pillar wording we agreed.
• One of the six is a comparison piece.
• I also write one placement-ready article for your PR agency to place on another site, or pitch two publications that AI tools already cite for your prompts.

Princeton and IIT Delhi found that adding specific statistics, quotations and named sources lifted visibility in generated answers by up to 34%, 40% and 27%. Those are ceilings from one study.

I use visible question-and-answer text because Ahrefs tested 1,885 pages that added schema markup, FAQ among the types, and found no measurable difference to AI citations.

A weekly cadence fits how fast citations fade: Stacker's tracking of more than 3 million citation events found a median half-life of 4.5 weeks for content on the wider web.

The off-site piece is in because brand mentions on other sites correlated with AI visibility at 0.66 to 0.71 in Ahrefs' study of 75,000 brands, against about 0.19 for the number of pages on a site. That is correlation, and it is the strongest pattern in the research I have read.

See 10 things to put in your article if you want ChatGPT to cite it and Why "more SEO" won't fix your AI search problem.

Checkpoint

• Four articles are live.
• I compare your weekly numbers with the baseline, platform by platform.
• I aim the last two articles at what has moved.

ChatGPT's citation rate swings independently of your content. seoClarity found citation volume fell 86% to 94% between February and April 2026 and recovered by May.

I read Perplexity and Gemini alongside ChatGPT, so one bad week does not decide the call. How long does AEO actually take? has the timeline.

Final review

• We review the data together.
• I report three numbers separately: how often AI names you, how often it recommends you and how often it cites your own pages.
• I add branded search and every answer from your enquiry form, and I refresh your first article.
• My recommendation on whether to move to the Retainer comes from that data.

The three numbers move independently. In Shero's study of 60 product categories, a recommended brand's own page was the cited source in only 31% of 159 cases.

One combined figure would hide which number needs work. Getting cited by AI search doesn't mean you're being recommended has the study.

Re-check

• Four weeks after the last article, I run the same prompts again.
• No new work goes in.

AI citations fade without new material, so the Week 12 numbers show how much of your result holds on its own.

That tells us whether you need the Retainer, using your own data.