I Rewrote the First 150 Words of My Top Service Page. AI Started Citing It.

Forty-four percent of all LLM citations come from the first 30% of a page's text. The lead is roughly twice as likely to be cited as the conclusion. That single data point is why I am rewriting the first 150 words of every money page on this site. Service pages, headshot verticals, top-of-funnel guides, all of it. The reason a corporate photographer ends up writing about copywriting is simple: a citable lead paragraph is what makes the rest of the page (the photographs, the schema, the multi-modal effort) get pulled into AI answers in the first place.

The order of operations is the same on every page: open it, read the first paragraph out loud, ask whether it actually answers the question my customer would type into ChatGPT, and if it doesn't, replace it with one that does. This is not a content refresh in the 2018 SEO sense. It is a deliberate restructuring of the lead so that an AI retrieval system, scanning for a self-contained block it can lift into an answer, finds one on the first scroll. Here is the data behind why and the framework I run on each page.

The data: 44% of citations come from the first 30% of your page

Growth Memo published an analysis in February 2026 of where inside a page LLMs pull citations from. 44% of all LLM citations come from the first 30% of a page's text, only 25% from the final third. The lead is roughly twice as likely to be cited as the conclusion. AI retrieval systems are scanning the top of the page for a chunk that fully answers the query, not reading your 2,000-word essay end to end.

The second number locked it in. 2026 AIO research from Conductor, Seer Interactive, and the Wellows/Ahrefs structured-data study converged on answer length: content that fully answers a question in a self-contained block of 134 to 167 words gets cited at roughly 4.2 times the rate of content that buries the answer or splits it across sections. AI systems retrieve at the paragraph level, not the page level. Chunk size matters.

I cover the broader signal mix in my AI search visibility pillar. These two numbers are why the lead-paragraph rewrite sits at the top of the priority list.

What "answer-first" actually means

Answer-first means the first paragraph of your page is the answer. Not the setup. Not the throat-clearing. The answer.

This cuts against twenty years of conversion copywriting habit. We were trained to open with empathy, build rapport, address the pain, then deliver the value. That structure still works for human readers in long-form sales pages. It is poison for AI retrieval. An AI scanning for a citable answer to "how much does corporate photography cost in St. Louis" does not care about your three-paragraph windup. It is looking for a paragraph with a number, a location, a service description, and a verifiable specific. If the first 167 words don't contain those, it moves on.

Answer-first does not mean cold or robotic. It means the answer leads. Personality, story, and nuance go in paragraph two, where they support the lead instead of delaying it.

The before-and-after: a corporate photography page lead

Here is a realistic before I see on nine out of ten service pages I audit. Passive, generic, zero specifics.

> "When it comes to corporate photography, there are many factors to consider. We've been serving businesses for years and offer a range of services tailored to your needs. Whether you're looking for headshots, lifestyle shots, or commercial work, we have you covered."

Four sins at once: throat-clearing ("when it comes to"), passive vagueness ("many factors," "a range of services"), zero verifiable specifics, and the low-trust catch-all "we have you covered."

The rewrite. 150 words, single paragraph, self-contained, with a location cluster, price range, turnaround time, and named industries and clients.

> "In St. Louis, Chicago, Dallas, and Kansas City, I photograph corporate teams of 25 to 50 people for $50 to $95 per person, including retouching and same-week delivery. Most programs run as a single onsite day per office, with proofs typically in your inbox the next business day and final retouched files within five business days. I work primarily with financial services firms, healthcare systems, professional services, and engineering and construction groups, including teams at Merrill Lynch, Northwestern Mutual, Washington University Medicine, and Wiegmann Associates. Every session uses the same calibrated lighting kit, gray-gradient backdrop, and retouching pipeline, so a new hire photographed eighteen months from now will match the original team page exactly. Standard deliverables include one polished headshot per person in three crops (square LinkedIn, wide bio, vertical print), plus a full-team group composite if your team page needs one. Send your headcount and target shoot week to get a quote."

Fifteen specifics in one paragraph: three cities, two price points, a team size range, two turnaround windows, four industries, three named clients, three crops, a next step. A prospect knows exactly what they would buy, for how much, on what timeline.

Entity density: the 15-per-1,000-words rule

The second rule is entity density. An entity is anything an AI can recognize and verify against its knowledge graph: named places (St. Louis, Chicago, Dallas), named companies (Merrill Lynch, Northwestern Mutual, Washington University Medicine), named industries (financial services, healthcare, legal), named regulations (FDA, HIPAA, GDPR, SOC 2), named institutions (WashU School of Medicine, Saint Louis University). Each is a node the AI can connect to, and each one increases its confidence that your page is talking about a real, verifiable thing.

The rule from 2026 AIO research is 15 or more recognized entities per 1,000 words. Below that, content reads as generic. At 15 and above, retrieval models start treating the page as authoritative on the topic cluster. It is a quantified version of the old advice to be specific instead of vague.

Low entity density: "We work with healthcare clients on team headshots." One entity, not citable.

High entity density: "We have photographed teams at Washington University School of Medicine, Saint Louis University, A.T. Still University, and the University of Missouri. We document for HIPAA-aware contexts and have shot for FDA-regulated medical device manufacturers running 510(k) submission documentation."

Nine entities. An AI retrieving on "medical device photographer" or "WashU healthcare team photographer" now has a verifiable source. Same length, different signal.

Practical move: every time you would write "healthcare clients," name three institutions. Every time you would write "the Midwest," name three cities. Every time you would write "various industries," name four.

Common opening mistakes that kill citation odds

Four mistakes I see on most service-page leads, in order of frequency:

1. Throat-clearing. "When it comes to," "in today's world," "now more than ever." These eat up your first 30 to 50 words without delivering a single specific. Cut on sight.

2. Passive voice. "Headshots are offered," "services are provided." Passive voice signals to AI retrieval that the page is talking about an abstraction rather than a real provider doing real work. Switch to active voice with a clear subject: "I photograph," "my team handles."

3. Over-promising without specifics. "World-class quality," "exceptional service," "unmatched results." Superlatives without evidence are a textbook low-quality marker. Replace each with a number or named outcome. "Industry-leading" becomes "4,000 professionals photographed." "Exceptional turnaround" becomes "proofs within one business day."

4. Generic claims. "We help businesses succeed." "We tell your story." The test: would the sentence still be true if you replaced your business name with a competitor's? If yes, delete it.

If your current lead contains any two of these four, the AI is almost certainly skipping it.

A 20-minute rewrite framework

The process I run on every page.

Minute 0 to 3: pick the page and the question. Open the page in one tab. Open ChatGPT or Perplexity in another and search the query you want it cited for. For my corporate photography page, the question was "how much does a corporate photographer cost for a team in St. Louis."

Minute 3 to 8: list the entities. Before writing a sentence, list what you will name: at least one location, one number, three named clients or industries, one named deliverable. The skeleton.

Minute 8 to 15: draft the paragraph. Sentence one: number and location. Two: turnaround or deliverable specific. Three: named industries or clients. Four: the standard you use. Five: deliverables. Six: call to action. Six sentences, roughly 150 words.

Minute 15 to 18: count words and entities. Under 134, expand the entity list. Over 167, cut adjectives. Aim for 8 to 10 entities inside the paragraph itself.

Minute 18 to 20: read it out loud. If it sounds robotic, soften two or three transitions. Lead with the answer. Personality belongs in paragraph two.

Where this fits in the order of operations

Order matters, because rewriting page leads on a site that AI bots cannot crawl is wasted work.

First, fix your robots.txt and crawler permissions. If GPTBot, ChatGPT-User, ClaudeBot, and PerplexityBot are blocked at the Cloudflare or robots.txt layer, no rewrite produces a citation. Setup is in my AI crawler robots.txt setup post. Do that first.

Second, ship FAQ schema. FAQPage JSON-LD is the single highest-impact structured data type for AI retrieval. Template is in my FAQ schema for ChatGPT post. Schema gives the AI a map of the page's Q-and-A structure; the rewritten lead gives it a citable paragraph at the top.

Third, do the answer-first rewrite. Once bots can read you and schema gives them structure, the lead becomes the unit retrieval systems are most likely to cite. Full sequence and the live experiment on this site are in the AI search visibility pillar.

What to do today

Three actions, no new tools or budget.

One: pick the page on your site that produces the most revenue. If unsure, pick the page you most want a prospect to land on after an AI search.

Two: read the first 167 words out loud. Count throat-clearing words, named entities, verifiable specifics. If named entities are below 8, or throat-clearing is above zero, the rewrite is your highest-return marketing task this month.

Three: run the 20-minute framework above. List entities, draft six sentences, count, read out loud, ship. Top three pages first is roughly 60 minutes of work covering most of your AI citation surface area.

If you want a second pair of eyes on your top page, get in touch and we'll talk through what I would change.