LLM Visibility Audit Scopes Compared for Lovable Sites

Gartner's forecast that traditional search engine volume would fall 25% by 2026 stopped being a prediction and became a budget problem. The buyers who used to start on a Google results page now start inside ChatGPT, Perplexity, or Gemini — and they form a shortlist before a single branded search happens.

That shift turned "llm visibility audit" into a procurement line item. It also turned it into a category where two vendors can quote the same three words and mean completely different work.

The gap matters more for Lovable sites than most. A React build that renders beautifully in a browser can hand an AI crawler an empty shell — and an audit that never opens that shell will happily grade the wrong thing.

What Does an LLM Visibility Audit Actually Measure?

A credible llm visibility audit measures three layers: whether AI crawlers can fetch and render the site's content, which real buyer prompts already surface competitors instead of the brand, and how clearly the site's entities and answers are structured for extraction. An audit reporting on fewer than all three is a partial audit, however polished the PDF.

Start with access. If GPTBot, PerplexityBot, OAI-SearchBot, and Googlebot can't retrieve rendered HTML, nothing downstream is real. OpenAI publishes its crawler behaviour and user-agent details in the OpenAI platform bot documentation, which is the fastest way to check whether a site's robots directives are quietly locking retrieval out.

Then citation footprint. This is the layer most audits actually sell: a fixed prompt set run across ChatGPT, Perplexity, Gemini, Google AI Overviews, and Bing AI Copilot, recording who gets mentioned and who gets linked. The useful version includes competitor names, not just the client's own share of voice.

Then entity and answer clarity — schema markup, answer capsule formatting, an llms.txt file, and consistent entity signals off-site. Google Search Central's guidance on AI features and helpful content is explicit that extractability and genuine usefulness drive inclusion, which is why adding schema markup to a Lovable website sits in this layer rather than the content layer.

Skip any one of the three and the audit still looks professional. It just isn't measuring the thing the buyer is worried about.

Why Does a Lovable Site Change What the Audit Must Cover?

Lovable ships React-based builds, so rendering configuration — prerendering, server-side rendering, or neither — determines whether AI crawlers see the content at all. Generic audits assume a server-rendered CMS. On a Lovable site, that assumption is the whole ballgame.

The scope document needs to name specific build-level items: Prerender.io or SSR setup, TanStack Start routing behaviour on dynamic routes, Supabase-driven pages that fetch content client-side, sitemap generation, and robots directives written for AI user agents specifically. The Lovable documentation covers publishing and prerendering options; what it can't do is force a vendor to put those checks in writing.

Performance and rendering behaviour are inseparable here. Guidance on rendering performance from web.dev explains how client-side hydration defers content availability — which is precisely the window where an AI crawler gives up.

Schema injection and llms.txt generation are engineering tasks, not copywriting tasks. If they aren't listed as deliverables, they aren't happening, no matter how many content recommendations the report contains.

In AIFun Agency's work with Lovable clients, the most common scope gap is an audit that scores content quality while never verifying what an AI crawler actually receives from the deployed build.

Here's what that gap looks like in practice on a single Supabase-backed Lovable page:

Before — a raw fetch with a GPTBot user agent returned the page shell, navigation, and a loading state. The pricing table, the case study copy, and the FAQ were all present in the browser and absent from the response body. The same copy that scored well in a human review was invisible to every crawler.

After — with prerendering enabled and the dynamic routes pre-generated at build time, the same fetch returned 1,400 words of body copy, structured headings, and JSON-LD. Nothing about the content changed. Everything about its retrievability did.

An audit that reads the page in a browser will never surface this. One that compares rendered output against raw fetch output finds it in under ten minutes.

How Do the Three Common Audit Scopes Compare?

The market mostly sells three scopes: a one-time visibility snapshot, a monthly prompt-tracking standard, and a full-stack AEO/GEO audit with an implementation roadmap. They differ less in effort than in what they leave the buyer holding.

1. The one-time snapshot. Small prompt set — often 10 to 25 questions — across one or two engines, usually ChatGPT and Google AI Overviews. Competitor benchmarking is shallow or absent. Fixes are described in general terms ("improve content depth"). No re-run is included, so the report is stale within a month. Priced as a fixed-fee diagnostic. Useful for a directional read before a bigger commitment; useless as a contract deliverable.

2. The monthly prompt-tracking standard. A larger prompt set, 50 to 150 questions, tracked across two to four engines on a recurring cadence, with competitor share-of-voice charts and citation-source breakdowns. The weakness is usually the fix layer: the report tells you Perplexity quotes a competitor's comparison page, then stops at "publish comparison content" without addressing whether your Lovable build renders it in a form Perplexity can parse. Priced as a retainer. This is the tier most vendors upsell and most buyers actually need — if the technical layer is bolted on.

3. The full-stack AEO/GEO audit. Prompt tracking plus crawl-and-render verification, schema and llms.txt review, entity consistency checks across Reddit, review sites, and directories, and a prioritised fix list mapped to the actual build — plus a scheduled re-run to prove movement. Fewer prompts than tier two sometimes, and that's fine. Depth beats volume. Priced as a project with an optional retainer attached.

The hidden variable across all three is engine coverage. A ChatGPT-only audit is not a GEO audit. Different engines retrieve differently, weight different sources differently, and cite differently — the same page can rank first in Gemini and be absent from Copilot for structural reasons an email-only report will never explain.

One more thing worth naming: tier two is where most contracts land, and tier two is where most disappointments originate. Monthly tracking without implementation guidance produces a very well-documented problem.

Which Line Items Should Never Be Optional?

Five items belong in any proposal worth signing, regardless of tier or price. Hold every vendor to all five.

• The prompt set, written out in full. Not "120 prompts across the buying journey." The actual questions, in the actual phrasing buyers use. If a vendor won't show them before signature, the prompt set is probably recycled across clients. - A methodology note on query fan-out. How were prompts derived from real buyer language — sales calls, Reddit threads, search console queries, support tickets? Query fan-out is where the analytical value lives, and it's the first thing a hollow audit skips. - A named engine list with model versions and test dates. LLM answers drift weekly. A report without a timestamp isn't a measurement, it's an anecdote. A report without model versions can't be reproduced. - A competitor set and citation-source breakdown. Who gets mentioned, and where those mentions originate: Reddit threads, G2 pages, listicles, documentation. That breakdown is the roadmap. Without it, the buyer knows they're losing but not where. - A prioritised fix list mapped to the Lovable build, plus a re-run date. Fixes should name the file, the component, or the build setting — not the content strategy. And the re-run should be priced or scheduled in advance, not negotiated later.

If a vendor delivers four of five, the missing one is usually the fix list. That's not an accident. The fix list is the part that requires knowing how Lovable actually builds.

What Red Flags Signal a Hollow or Inflated Scope?

A proprietary "AI visibility score" with no published formula is the clearest warning sign, followed closely by deliverables made of screenshots with no prompt text or timestamp. Both let a vendor show motion without ever showing evidence.

Watch for the absence of a competitor benchmark. Visibility in a vacuum is meaningless — the question is never whether the brand gets cited, it's whether it gets cited instead of the three companies it loses deals to.

Content-only scope on a Lovable site is a structural red flag. If the audit recommends publishing comparison pages and FAQ blocks but contains no rendering, prerendering, or schema work, those recommendations cannot land. The fixes will be written, the content will be published, and the crawlers will still receive a shell.

Then there's scope creep in reverse: an audit priced cheaply because implementation is billed separately, at a premium, on a proposal that hasn't been written yet. An audit whose entire value is the fix list shouldn't treat the fix list as an upsell.

Finally, look at re-measurement. No re-run window means no accountability. A vendor confident in the diagnosis will schedule the proof.

Price is the last thing to compare, not the first. Two quotes 40% apart usually differ by a single line item — the render verification or the re-run — rather than by the quality of the analysis. Ask each vendor to price the missing item separately. The gap often closes, and when it doesn't, the cheaper proposal was never offering the same work.

How Should You Compare Two Quotes Side by Side?

Build a six-row matrix — engines, prompts, competitors, technical checks, fix list, re-run cadence — and score both proposals in the same sitting. The exercise takes twenty minutes and exposes most of the difference.

Weight technical checks higher than raw prompt volume for a Lovable site. A 300-prompt audit that never verifies rendered fetch output is worth less than an 80-prompt audit that does, because the first one measures a page nobody can retrieve.

The matrix should also record what each vendor hands over in writing. A proposal that promises a live dashboard but no static report leaves the buyer with nothing to file, nothing to share with a board, and nothing to compare against the next quarter's run. Ask whether raw response transcripts are included, whether the prompt set arrives as a spreadsheet, and whether the fix list is formatted as tickets a developer can pick up. Those artifacts cost the vendor almost nothing to produce, and they separate a working document from a sales asset.

Then run a live test on the call. Give both vendors the same prompt — one your buyers actually type — and ask each to run it now, in front of you, and explain what they see. The vendor who can't do it live is working from a template.

Ask one more question: what would you fix first? A specific answer referencing prerendering, JSON-LD, or a citation gap on a named source is a good sign. A general answer about content authority is not.

For teams that would rather not run the comparison at all, the alternative is handing the entire scope to a Lovable specialist. That's a legitimate choice — provided the scope document still contains the five non-negotiables above.

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.

The Scope Document Is the Only Part of the Contract That Matters

Every audit sold in 2026 will arrive as a polished PDF with charts. The charts are not the deliverable.

The deliverable is a named engine list, an explicit prompt set, a competitor benchmark, a crawl-and-render verification of the deployed Lovable build, a prioritised fix list mapped to code and configuration, and a date when all of it gets re-measured. Everything else is presentation.

Buyers who hold vendors to those six items get audits that change rankings. Buyers who accept a visibility score and a screenshot deck get a document that confirms a problem they already suspected. The difference isn't budget — the cheapest tier can include the fix list if the vendor knows how to build one. For a broader view of what crawler-facing configuration looks like on this stack, the Lovable SEO checklist for AI crawlers and the guide to getting a Lovable site cited by Perplexity cover the technical ground most scope documents skip.

Sign the scope, not the pitch deck.

Ready to get cited across AI search without doing it yourself? Start with the AIFun AEO / GEO Dominator → https://aifunn.com