The DNA of a Prompt 2/10: The Title Is Not Bait, It Is an Address

Everyone writes titles thinking about who will click. That makes sense, because it is the only part we ever see working.

But the title works long before that, in a place nobody looks, and there it decides whether your text gets considered at all.

What the title does before anyone reads it

When a system cuts a paragraph out of the middle of your page, it needs to know what that paragraph belongs to.

Patent US 9,959,315 B1, from Google, describes how this gets resolved. It was filed on January 31, 2014 and granted on May 1, 2018. The system builds a path through the heading hierarchy, from the root heading down to the heading the passage sits under. That path carries the text of every level it passes through. ^{1 2}

Translated to your screen: the paragraph inherits the headings above it. All of them.

A passage about pricing, written under “Consulting for corporate architecture”, is one thing. The same passage, under a generic heading like “Information”, is something else, and far harder to locate.

This is called the heading path.

The title keeps attracting clicks. That part does not go away. What changes is the order of importance: before it attracts anyone, it has to locate the passage.

Two components that depend directly on the title

The same patent describes components of the context adjustment, and two of them depend on what you write in the heading. ³

The first is the depth of the heading inside the hierarchy. Where it sits in the tree matters.

The second is the similarity between the question text and the heading text. Here is the most direct practical consequence in this article.

If someone asks “how much does SEO consulting cost” and your heading is “Investment”, the textual similarity is zero. The heading may be elegant, sober, on brand. For that component, it does not exist.

A heading without the words of the question loses similarity. However much better it sounds.

The ruler, before the rules

Before any writing instruction, one boundary, and it holds for all ten articles in this series.

No writing requirement guarantees citation. What these requirements do is raise the probability that the passage gets retrieved and attributed to you. ⁴

Being retrievable is a necessary condition. It is not a sufficient one.

I put this here, before the rules, and not in a footer after them. A ruler that shows up only at the end becomes fine print, and fine print exists to go unread.

What Google says you do not need to do

The industry sells a list of prerequisites for “showing up in the AIs”. A text file for robots, special markup, a Markdown version of the site.

The official Google documentation says the opposite. The page “AI features and your website”, updated on December 10, 2025 and checked on August 1, 2026, states that you do not need to create machine-readable files, AI text files, markup, or Markdown to appear in Search. ⁵

I keep this in a piece about titles for a reason. People buy a solution to a problem the documentation says does not exist. And the thing that actually weighs, the heading text, sits there with nobody tending it.

The 150-character ceiling

There is a published number that changes the way you write, and it does not come from Google.

In Anthropic’s public documentation on web search, the cited_text field is capped at 150 characters of the passage actually used. ⁶

That is shorter than this paragraph.

It turns into a design constraint. The sentence that answers has to fit that space on its own, with the subject inside it. If the answer starts with “this happens because”, the cut carries the explanation without carrying the subject, and without carrying you.

On formats, there is experimental evidence. The paper “GEO: Generative Engine Optimization”, by Pranjal Aggarwal and others, published on arXiv under identifier 2311.09735 on November 16, 2023 and accepted at KDD 2024, tested variations on a benchmark of 10,000 queries over nine datasets. ⁷

The prompt

This is the titles prompt of the series. It works the title as an address, and not as bait.

You are a technical content editor.

I will give you the subject of a section and the question it answers.
Write five possible headings for it.

For each heading, report:
1. how many words from the question appear in it
2. the full path, from the root heading down to it
3. whether the first paragraph under that heading would answer
   the question on its own, without leaning on the previous one

Rules:
- the subject of the section appears inside the heading
- nothing starting with "Understanding", "Getting to know" or "Learn more"
- no promise of results and no superlatives
- the heading describes what the section delivers, not what
  the reader is going to feel

Question the section answers: [question]
Subject: [subject]

What it does: it forces you to test the heading against the question, and it returns the path explicitly.

What it does not do: it promises no click, no position, and no citation. It compares alternatives against a criterion.

Frequently asked questions

Do longer or shorter titles perform better?

The patent says nothing about length. It describes depth in the hierarchy and similarity with the question text. A short title carrying the words of the question serves both better than a long, vague one. ³

Do I need Markdown or an AI file on my site?

According to the official Google documentation, updated in December 2025, no. You do not need to create machine-readable files, AI text files, markup, or Markdown to appear in Search. ⁵

Does a title in question form work?

It tends to raise similarity with the query, which is one of the described components. And the patent also mentions detecting a question immediately before the passage as a distinctive trait. Those are two different effects, and neither one is a guarantee. ^{3 4}

Notes and sources

1. Granted patent US 9,959,315 B1, “Context scoring adjustments for answer passages”, Google LLC. Priority and filing on January 31, 2014, grant on May 1, 2018, status active. The mechanism itself is detailed in piece 1 of this series.
2. Literal definition of the heading path in the same patent: the vector runs from the root heading to the heading the passage sits under, carrying the text of every level.
3. Components of the context adjustment in the same patent: depth of the heading in the hierarchy, similarity between the question text and the heading text, coverage ratio of the passage, and detection of distinctive textual traits, such as a question immediately before the passage or list format.
4. Declared boundary: no writing requirement guarantees citation. Such requirements raise the probability that the passage gets retrieved and attributed. Being retrievable is a necessary condition and not a sufficient one.
5. Google Search Central, “AI features and your website”, updated on December 10, 2025, verified on August 1, 2026. Literal statements reproduced in the body.
6. Anthropic public documentation on web search: the cited_text field is capped at 150 characters of the passage actually used.
7. Pranjal Aggarwal, Vishvak Murahari, Tanmay Rajpurohit, Ashwin Kalyan, Karthik Narasimhan and Ameet Deshpande, “GEO: Generative Engine Optimization”, arXiv 2311.09735, first version on November 16, 2023, accepted at KDD 2024. Benchmark of 10,000 queries over nine datasets.