Last updated: July 13, 2026
The GEO playbook is nine steps ordered by lift per hour. Do the cheap, high-lift moves first: open technical access, write answer-first, add facts, add expert quotes, then build entity and off-domain signals. This post walks each step in order, shows why it works, and gives you the example and the mistake to avoid at each one. Read our full complete GEO guide first if you need the definitions, then come back here to execute.
Order is the difference between a playbook and a wish list. Every step below earns citations, but they do not cost the same time, and they do not unblock each other equally. Technical access is free and gates everything else. Answer-first writing is a habit, not a project. Off-domain mentions cost outreach hours and only pay off once engines already trust your on-site signals.
Rank the work by expected citation lift divided by the hours it takes. Cheap wins go first so you see movement fast and fund the expensive steps with proof. If you are new to the field, start with what generative engine optimization is before you sequence your own sprint.
Allow AI crawlers in robots.txt and add an llms.txt file. Nothing else works if engines cannot read you.
If a crawler cannot fetch your page, no passage on it can ever appear in an answer. Citation begins at fetch time, not at publish time. Block the bots and every later step in this playbook is wasted effort.
Open your robots.txt and allow GPTBot, PerplexityBot, ClaudeBot, and Google-Extended. Then add an llms.txt file at your site root that lists the pages you want models to read and the ones you prefer they skip. Anthropic proposed llms.txt in September 2024 as a machine-readable map for AI agents. Keep the file short and current.
A SaaS site we audited had a security plugin that defaulted to blocking all bots not on a legacy allowlist. The team ranked on Google but never appeared in ChatGPT answers. One robots.txt change and an llms.txt map moved them into retrieval within two weeks.
Teams set a crawler rule once and forget it. A later plugin update, a CDN setting, or a staging redirect can reblock the bots silently. Check the file quarterly and after any platform migration.
Open every section with a direct answer in the first 40 to 60 words. AI extracts passages, not pages, so the answer has to stand alone at the top.
Models pull a sentence or a short block to satisfy a query. If your real point sits three paragraphs down, the engine quotes a competitor whose answer is upfront. Wellows found engines prefer passages of 134 to 167 words, so keep each answer block tight and self-contained.
State the answer in the first sentence of every H2 and H3. Then support it with evidence. Match the phrasing people actually use in prompts, and keep each idea in its own short paragraph so the model can lift one block without dragging in unrelated context.
Instead of opening a section with “There are many factors to consider when thinking about citation,” write “Citations rise when you name a source, add a statistic, and quote an expert. The Princeton study measured a 40 percent lift from added sources alone.” The second version is extractable on its own.
Writers bury the answer behind a story or a company intro. The model has no patience for your origin story. Lead with the payoff, then earn the right to add context.
Attach a statistic and a source to every claim. The Princeton study found citations with sources lift visibility by about 40 percent, and statistics by 37 percent.
Numbers are the cheapest citation win you have. The Princeton and KDD 2024 study by Aggarwal et al. measured that adding sources lifted citations by 40 percent, adding statistics by 37 percent, and adding expert quotations by 30 percent. Named, dated facts give a model something concrete to repeat.
Put one statistic and one source name next to each key claim. Date the study, name the authors or organization, and link to the original. Ahrefs found that 76 percent of links cited in Google AI Overviews come from the top 10 organic results, so pair your facts with pages that already rank.
A pricing page claimed “most teams overpay for tools.” We changed it to “most teams overpay: 76 percent of AI Overview citations come from the top 10 organic results, per Ahrefs, so weak rankings quietly cap your visibility.” Citation requests for that page rose the next month.
Teams cite a stat without naming the source or the date. A bare number looks like filler to a model and to a reader. Always anchor the fact to a named, findable origin.
Add a named quote from a recognized expert, with title and affiliation. The Princeton study measured a 30 percent lift from quotations.
A named expert quote gives the model a verbatim line it can trust and attribute. The Aggarwal et al. study found expert quotations lifted citations by 30 percent. Quotes also satisfy the experience and authority signals that search and answer engines both reward. See our take on why experience beats keywords for E-E-A-T.
Find a practitioner or researcher with a real title and affiliation. Ask one sharp question, publish their answer with their name attached, and keep the quote short enough to extract whole. One quote per pillar page is enough to move the needle.
On a fintech explainer we added a two sentence quote from a named compliance officer. That single attributed line started appearing in Perplexity answers about the topic, because the model preferred a person’s wording over our generic definition.
Quotes from “a spokesperson” or an unnamed “industry expert” carry no weight. If you cannot name the person and their role, the lift does not happen.
Mark your brand and authors with Organization and Person schema, claim a Wikidata entry, and align your sameAs profiles. Engines treat you as a known entity instead of a random page.
Structured data tells a model who you are and how your pages connect. SEOcrawl found that 71 percent of pages cited by ChatGPT include structured data. Entity clarity is the gap between being cited once and being cited consistently across a topic.
Add Organization, WebSite, and Person schema to your templates. Claim or correct a Wikidata entry for your brand, and make your sameAs profiles point to the same handles everywhere. Consistency across properties is what lets a model resolve “you” across the web.
A consultancy had three different founder bios across their site, LinkedIn, and a podcast page. We aligned the name, title, and photo, added Person schema, and corrected their Wikidata record. Within a quarter, ChatGPT began attributing quotes to the founder by name.
Teams add schema once and let author bylines drift. When the name on the page no longer matches the name in the graph, the entity signal weakens. Keep bios identical everywhere.
More than 90 percent of citations come from third-party sources. Show up on Reddit, LinkedIn, and in press where AI actually looks.
Your own site is necessary but not sufficient. More than 90 percent of AI citations come from third-party sources, not a brand’s own domain. AuthorityTech found Reddit makes up about 46.5 percent of Perplexity citations, and Bing retrieval overlaps ChatGPT retrieval by about 87 percent, so off-domain presence feeds multiple engines at once.
Publish useful answers where the conversations already happen. Join relevant Reddit threads with real help, post practitioner takes on LinkedIn, and pitch reporters who cover your space. The goal is presence in the sources models pull from, not vanity traffic.
A developer tool answered questions in two niche subreddits with no links, just solutions. Those threads became Perplexity citations. The brand never ranked number one for the phrase, yet still got named in the answer.
Teams assume ranking number one on Google guarantees AI citations. Shadow found only about 11 to 12 percent of pages ranking number one on Google are cited by ChatGPT. Owned rankings do not equal answer presence.
Cover a topic fully with a pillar and supporting posts, linked internally. AI favors comprehensive sources it can trust across a subject.
One isolated page rarely wins. A cluster with clear hierarchy tells the engine which page owns the intent and lets it cite you across related questions. Depth signals that you are the source, not a single lucky article. Build this with a real topical authority cluster rather than scattered posts.
Pick one pillar topic and write the supporting pieces that answer every adjacent question. Link them in a clear hub and spoke pattern. Each post should be extractable on its own and should point back to the pillar for context.
A health brand built one pillar on “intermittent fasting” plus nine supporting posts on schedules, risks, and myths. Perplexity began citing different posts from the cluster for different subquestions, and the pillar rose as the trusted hub.
Teams publish ten unrelated articles and call it depth. Without internal links and a shared schema, the engine sees noise, not a cluster. Link the pieces together on purpose.
Update key pages within 30 days. Wellows found content refreshed at 30 days earns 3.2 times more citations than stale content.
Recency is a ranking and citation signal at the same time. Wellows found that content refreshed within 30 days earns 3.2 times more citations than stale content, and that 95 percent of ChatGPT citations come from sources under 10 months old. Freshness is not optional polish. It is a core requirement.
Mark your top pages with a refresh date and review them on a 30 day cycle. Update statistics, add a new expert line, and confirm links still resolve. Treat freshness as a recurring task, not a one time launch.
A comparison page we maintained got a quarterly stats update and a new expert quote every month. Its citation rate stayed high while competitors’ older pages slid out of the answer set. The 3.2 times lift tracked exactly with the refresh habit.
Teams set a publish date and walk away. A page that was great at launch becomes stale the moment a newer source appears. Build the reminder before you need it.
Run a 50-prompt baseline, then track citation rate monthly. Change the pages that are not in the retrieval set, and double down on what gets cited.
GEO is a feedback loop. The teams that win are the ones that measure, not the ones that publish and forget. A 50-prompt baseline tells you where you already appear and where a competitor owns the answer you wanted.
List 50 queries your buyers ask. Run each through ChatGPT, Perplexity, and Google AI Overviews. Record whether you are cited, mentioned, or absent. Repeat monthly and act on the gaps. This is the same discipline we cover in how to get cited by ChatGPT, Perplexity, and AI Overviews.
A B2B tool found it was cited for “best X software” but absent for “X vs Y.” We built the comparison content and earned the citation within one cycle. Measurement turned a blind spot into a win.
Teams track rankings and call it GEO measurement. Rankings and citations diverge. You must measure answers directly, not assume a number one rank means an answer mention.
Start with the zero cost moves that gate everything else. One person can open technical access and rewrite existing pages answer-first in the first week. Facts, quotes, and schema are edits to content you already have, so they fit inside normal publishing.
Off-domain mentions and topical depth need the most hours, so schedule them after the on-site base is solid. A two person team should assign one owner to on-site signals and one to off-domain presence, then meet weekly on the measurement sheet. This keeps lift per hour high without burning the small team on low return work.
Give every step a number you can watch. Technical access is binary: are the bots allowed. Answer-first is a yes or no pass on each section. Facts and quotes are a count per page. Entity authority is a schema check plus a Wikidata status.
Off-domain mentions are a tally of citations from third party sources, which is where the 90 percent figure matters most. Freshness is days since last update. Measurement is your monthly citation rate on the 50 prompt sheet. When a step has no number, it is not done, it is just hoped for.
Week one opens access and rewrites the top five pages answer-first. Week two adds facts, sources, and one expert quote to each. Week three ships schema, claims the Wikidata entry, and begins off-domain answers on one community. Week four runs the 50 prompt baseline and sets the 30 day refresh calendar.
This cadence fits a small team because each week builds on the last. By day 28 you have a measurable baseline and a living process, not a document that sits in a drawer. Extend the sprint into a second month by repeating weeks two through four on the next set of pages.
Technical access needs only your robots.txt editor and a text file for llms.txt. Answer-first writing needs a word counter and a discipline, not software. Facts and quotes need a citation manager and a spreadsheet of studies, starting with the Princeton and KDD 2024 data from Aggarwal et al.
Entity work benefits from a schema validator and a Wikidata account. Off-domain tracking is manual at first, then faster with a mention monitor. Measurement wants a prompt tracker you actually open each month. For a fuller list, see our best GEO tools for AI visibility. Pick tools that speed the steps above, not tools that replace the writing.
Print this and work top to bottom.
How fast does the playbook show results?
Baseline in week one. Early citation movement in 30 to 90 days as freshness and entity signals build. The 30 day refresh rule from Wellows is the lever that compounds first.
Do these steps guarantee a ranking?
No. They raise the odds of being cited. Consistency and off-domain presence do the rest. Shadow found only 11 to 12 percent of number one Google pages get cited by ChatGPT, so owned rank is never the whole story.
Which three steps first if I am short on time?
Open technical access, write answer-first, and add facts with sources. Highest lift, lowest effort. The Princeton study shows sources and statistics together deliver the largest early citation gains.
Is the checklist enough without tools?
Yes to start. Tools speed up tracking, but the nine steps work with manual prompt audits. A spreadsheet and a calendar cover most teams through the first two months.
Why does my number one Google ranking not show up in ChatGPT?
Ranking and citation are different systems. More than 90 percent of AI citations come from third-party sources, and only about 11 to 12 percent of top ranked pages get cited. You need off-domain presence and structure, not just rank.
How many prompts should I track for measurement?
Fifty is the floor. It gives enough coverage to see trends without becoming a chore. Expand to 100 once the cadence is a habit.
One short, useful email each month on SEO, copywriting, and GEO for B2B. No fluff, no spam.
The best GEO tools track your brand's visibility inside AI answers and, in some cases, ship the fixes for you. Google Search Console does not report AI citations, so a dedicated tool is the only way t
Read More
Generative Engine Optimization (GEO) is the practice of structuring your content and brand so AI search engines like ChatGPT, Perplexity, and Google AI Overviews cite you as a source. It does not chas
Read More
A real GEO service does more than edit blog posts. It runs a prompt baseline, restructures content for extractability, builds entity signals, earns off-domain mentions, and tracks citations month over
Read More