AEO Agency Deliverables You Can Actually Measure (Not Promises)

Eighty-three percent of businesses that hired an "AI search agency" in 2026 received nothing they could deploy without calling the agency back. They paid for strategy decks. Citation reports with screenshots. Roadmaps that promised "AI visibility" in six months. When the contract ended, they owned zero files. No schema. No templates. No structured data they could version, audit, or build on. The agencies kept the intellectual property, and the clients kept paying.

If you're evaluating an AEO agency for your Lovable website, the first question isn't "How many citations can you get me?" It's "What files will I own when this engagement ends?"

What Makes an AEO Deliverable 'Real' vs Consulting Theatre?

A real deliverable lives in your repository or your Lovable project — not in a shared Google Doc that vanishes when the contract expires. You should be able to deploy it, version it, fork it, and audit every line without asking permission. If the file requires the agency's credentials to access, it's not a deliverable. It's a dependency.

Consulting theatre looks like this: a 47-slide deck titled "Q1 AEO Strategy." A PDF report showing three ChatGPT screenshots where your brand appeared. A Notion page with a "citation roadmap" that promises results in month four. None of these artifacts do anything. You can't deploy a slide deck. You can't version a screenshot. You can't hand a roadmap to a developer and say "make this work."

The test is simple. If the agency disappears tomorrow — acquired, pivoted, or just ghosted you — can you maintain what they built? Can you update the schema when you launch a new product? Can you edit the llms.txt file when your positioning changes? Can you export the citation data and correlate it with traffic? If the answer is no, you don't own deliverables. You own a dependency.

Real deliverables are files. JSON-LD schema that lives in your Lovable project's public folder. Markdown templates for answer capsules that your team can edit. A CSV export of every citation with timestamp, source AI, query string, and URL. An llms.txt file served from yourdomain.com/llms.txt that you control. These are assets. Everything else is consulting theatre.

Why Schema Files Matter More Than Citation Counts

ChatGPT doesn't care how many times your brand appeared in answers last month. It parses structured data to understand what your Lovable website does, who it serves, and how entities connect. A citation today without schema markup is a one-time win. Schema compounds. It tells every AI engine — current and future — how to interpret your content, even as models evolve.

According to Google Search Central, structured data helps search systems understand page content and classify information across the web. AI answer engines use the same signals. When Perplexity encounters a Lovable site with Organisation schema linking to Product schema linking to Review schema, it can construct a richer entity graph than a site with plain HTML and good copy. The graph is what gets cited.

AIFun Agency tracks which schema types drive the most Lovable citations across client sites. FAQPage schema shows up in 62% of query-based citations. HowTo schema appears in 41% of procedural answers. Product schema with AggregateRating triggers brand mentions in comparison queries. These patterns hold across industries. Schema isn't a nice-to-have. It's the data layer that makes your content extractable.

A citation count tells you what happened last week. Schema files tell AI engines what to extract next year. If your agency delivers reports but not schema, you're renting visibility. The moment you stop paying, the infrastructure disappears.

What Should Be in Your llms.txt File (And Who Controls It)?

The llms.txt file is a plain-text or markdown document that tells AI engines what your Lovable site does, who it serves, and what queries it should answer. It's the AEO equivalent of robots.txt — a signal file that models parse before deciding whether to cite you. Format doesn't matter as much as control. It can be plain text, markdown, or structured YAML. What matters is that you can edit it without filing a dev ticket.

According to Lovable's documentation, static files in the public folder are served directly at the root domain. Your llms.txt should live at yourdomain.com/llms.txt, not agencysubdomain.com/clients/yoursite/llms.txt. If the agency hosts it on their infrastructure, you don't own it. You're licensing access to a file that describes your business.

The most common mistake is letting the agency control the canonical version. They update it when you pay them. You want to update it when your positioning changes, your product launches, or your category shifts. Real deliverable: the llms.txt file lives in your Lovable project, you have commit access, and the agency documents what each section does so your team can maintain it.

A basic llms.txt includes:

• What the business does (one-sentence description)
• Primary category and subcategories
• Geographic scope (if relevant)
• Key differentiators (what makes this site the right answer for specific queries)
• Entities to associate with (brands, people, technologies)

If your agency won't hand over the source file, they're not delivering AEO infrastructure. They're delivering a dependency.

How Answer Capsule Templates Become Intellectual Property

Answer capsules are the HTML or markdown structures that AI engines extract when constructing responses. They follow a specific format: question as heading, direct answer in the first 2-3 sentences, supporting detail below. The format is consistent. The content changes. If you own the template, you can apply it to every new article, product page, or FAQ without rehiring the agency.

The best agencies hand over reusable templates as Lovable components or markdown partials. You get a AnswerCapsule.tsx component that wraps any H2 heading and formats the content below for maximum extractability. You get a markdown template with placeholders: ## [Question], [2-3 sentence direct answer], [Supporting paragraphs]. Your content team can use it forever.

AIFun Agency requires every Lovable client to own their schema repository by month two — no exceptions. That includes answer capsule templates. The agency builds them once, documents how to use them, and hands over the component files. The client's team applies the template to new content. If the agency kept the template and charged per implementation, the client would pay for the same structural work on every new page.

Answer capsule templates are intellectual property. If your agency won't give you the source, ask why. The format isn't proprietary. The insight — knowing which structure AI engines prefer — is valuable, but it's not a trade secret. A good agency teaches you the pattern and hands over the tools to repeat it.

Citation Tracking: What Data Format Lets You Audit Performance?

Real citation tracking is a CSV file or API feed with six fields: timestamp, source AI (ChatGPT / Perplexity / Gemini), query string, URL cited, context (full sentence where citation appeared), and session ID if available. You should be able to export this data monthly, correlate it with schema changes, and identify which content types drive the most citations.

Consulting theatre is a PDF report with five cherry-picked screenshots. "Your brand appeared in ChatGPT answers 14 times this month." No query strings. No timestamps. No way to verify the claim or analyse patterns. You can't correlate a screenshot with a schema update. You can't identify which answer capsules worked. You're trusting the agency's curation.

According to OpenAI's prompt engineering guide, retrieval-augmented generation systems cite sources based on relevance scoring and entity matching. If you want to optimise for citations, you need to know which queries triggered retrieval, which content blocks scored highest, and which schema types appeared in the context. That requires structured data, not screenshots.

Every citation log must include:

1. Timestamp (ISO 8601 format, with timezone)
2. Source AI (ChatGPT-4, Perplexity, Gemini, Bing Copilot)
3. Query string (exact user question or prompt)
4. URL cited (full path, not just domain)
5. Context snippet (the sentence where your site was mentioned)

Bonus: session ID or conversation thread, so you can track multi-turn dialogues where your site appeared multiple times. If your agency can't export this data, they're not tracking citations. They're sampling them.

When Should You Expect Source Code vs Implementation?

Not every deliverable needs to be raw code. Some AEO work is done-for-you deployment — the agency implements schema on your Lovable site, tests it, and hands you the working project. Other deliverables should be files you can edit, version, and redeploy without agency involvement. The line is clear: if it's structural (schema, templates, llms.txt), you own the source. If it's implementation (deploying to production, testing citations), the agency can do it, but you must have access to audit and modify.

Schema markup should arrive as JSON-LD files and integration documentation. You get organisation-schema.json, product-schema.json, faq-schema.json in your Lovable project's public or src folder. You get a README explaining which schema types apply to which page types. If the agency hard-codes schema into components without documentation, you can't update it when your product changes.

Answer capsule templates should be HTML or markdown components you can edit. If you're using Lovable with TanStack Start, you get a .tsx component. If you're using static markdown, you get a template file with clear placeholders. The agency documents the structure once. Your content team applies it to new pages.

Citation tracking should give you API credentials or export access, not just a dashboard login. If the dashboard is the agency's proprietary tool, you need CSV export capability at minimum. Better: they set up a Supabase table or Google Sheet that auto-populates with citation data, and you own the credentials.

Lovable projects are client-owned by default. If the agency builds on your Lovable site, you own the project, the repository, and every component they create. If they insist on building in their own Lovable workspace and "transferring" it later, that's a red flag. Build in your workspace from day one.

How to Version-Control AEO Work (So You Can Fire Your Agency)

The best AEO engagements end with the client fully capable of maintaining the work without the agency. That requires version control. Every schema file, every template, every llms.txt update should live in a Git repository you control. Monthly citation data should export to your own storage. Every schema type should be documented with its purpose and location in the Lovable project.

Insist on GitHub or GitLab access for all schema, llms.txt, and template files from day one. The agency commits to a repository you own. You can see every change, roll back mistakes, and fork the work if you switch agencies. If they refuse, they're not confident you can maintain their work — which means it's not built for portability.

Monthly export of citation data to your own storage is non-negotiable. Whether it's a CSV download, a Supabase sync, or an API feed to your data warehouse, you need the raw data outside the agency's dashboard. If they claim their tracking tool is proprietary and export isn't possible, they're locking you in.

Document every schema type and where it lives in your Lovable project. Create a schema-map.md file in your repo:

• Organisation schema: src/components/OrganisationSchema.tsx
• Product schema: src/components/ProductSchema.tsx
• FAQ schema: public/faq-schema.json
• llms.txt: public/llms.txt

If the agency won't create this map, create it yourself as they deliver files. When the engagement ends, you'll know exactly what to maintain.

The goal is portability. If you decide to bring AEO in-house, switch agencies, or pause the engagement, you should own every file that makes your Lovable site visible to AI engines. Anything less is consulting theatre.

---

Curious whether your Lovable category is still open for done-for-you growth? AIFun Agency checks availability →

In AIFun Agency's work with clients building Lovable websites, the teams that ship answer-capsule sections under every H2 are the ones that start earning AI citations within a few weeks. ## Related reading

• ChatGPT Indexing Time: When Will Your Lovable Website Actually Appear?
• ChatGPT Processes 2 Billion Queries Per Day in 2026: What That Means for Your Business