The $4,200 AEO Retainer, Line by Line: What Holds Up and What’s Padding

The strangest number in groas’s edge logs is 1,071 requests for robots.txt in live-answer traffic, compared with 70 for a service page. That is why I chose this $4,200-a-month AEO proposal for a line-by-line teardown. It is a composite of four proposals I have seen, and its eight familiar charges let me ask a less glamorous question than “How visible are we in AI?”: What work gives an answer engine something useful to fetch?

My verdict: roughly half the lines are monitoring, reporting, or rituals no engine is documented to reward. The lines that hold up are crawl access, dated pages that answer specific buyer questions, and third-party mentions. The logs help order that work; they do not, on their own, prove that a fetch became a citation. Composite retainers like this can run from $3,000 to $25,000-plus a month, depending on the workstreams staffed. At $4,200, I want more than an attractive way to watch a chart move.

The eight lines on the proposal

The package starts with a visibility audit across ChatGPT, Perplexity, Claude, AI Overviews, and Copilot. Foundational fixes follow, then content and citation work, then ongoing tracking. That four-phase shape is common, and early mentions can swing for a week or two before settling over two to three months. The sequence sounds sensible. The billing deserves a closer look.

1. Monthly AI visibility score and share-of-voice dashboard
2. Prompt tracking across 50 to 100 prompts
3. llms.txt and other “AI files”
4. Schema sprint across key templates
5. Robots and crawl access for AI bots
6. Entity optimization and content refresh
7. New question-led content for buyer prompts
8. Citations and third-party mentions

I will take them in that order. A deliverable is not automatically a result, and a useful one-time fix is not automatically worth a monthly line item.

Line 1: The visibility score is a report, not the work

The first deliverable is one monthly number that claims to capture your share of AI answers. I used to pay attention to scores like this until I ran three trackers side by side for the same brand in the same week and got three different answers. That disagreement is structural: licensed panels, clickstream data, scraped chat interfaces, and server logs measure different things. You are buying one vendor’s definition of visibility, not visibility itself.

The underlying feed can also be inexpensive. Entry prices listed for these tools include $99 a month for Profound on one engine, $99 for Semrush on 25 prompts, and $29 for Otterly across four engines. That does not make an analyst’s interpretation worthless. It does mean an agency cannot wrap a feed in a $4,200 retainer and call the screenshot month-one progress.

Keep the baseline audit. Refuse the monthly score as a substitute for work.

Line 2: Prompt tracking earns its fee only when it changes the plan

Next comes tracking across 50 to 100 prompts. It sounds comprehensive until you run the same prompt again and see different sources. One methods study needed 33 to 94 answers before cited-site rankings settled; single runs produced different brand lists more than 99% of the time. A single pass across a long prompt list can dress noise as coverage.

I would rather track 10 to 20 buyer prompts, run each 3 to 5 times, and record share cited. Even that is a working signal, not a settled ranking. Its value is the next decision: Which question has no good page? Which existing page needs a clearer answer? If the proposal names neither a fixed prompt set nor a repeat count, the tracking is unlikely to connect cleanly to what gets published.

Keep prompt tracking only when each cycle ends with pages to write or fix. Otherwise, line 2 is line 1 with more rows.

Line 3: llms.txt is an easy file to sell and a weak fee to defend

The proposal calls llms.txt a translation layer for AI models and bundles it with other “AI files.” I have asked three agencies what citation lift the file produced. None had a number. Google Search says it neither helps nor harms visibility; the findings cited in the draft also put AI-retrieval requests for these files at the margins. That is a thin basis for a billed workstream.

Could someone create the file? Sure. But the question on this proposal is whether creating it deserves its own monthly charge while buyer-question pages wait to be written. Let the agency ship llms.txt if it insists; strike it as a recurring line item. I would ask for evidence of a benefit before paying for the ritual again.

Line 4: Schema is housekeeping, not a citation sprint

Schema is the line I argued about with myself longest because I used to sell it as citation work. I would now call it hygiene. In one comparison, 1,885 pages that added JSON-LD were measured against around 4,000 controls; AI Mode rose 2.4%, ChatGPT rose 2.2%, and AI Overviews fell 4.6%. Those mixed results do not make a convincing case for a monthly citation lever. In a separate test described alongside that analysis, eight prices appeared only in JSON-LD: Claude recovered none, Perplexity one, and Gemini half.

The lesson is not to remove valid schema. Keep it for search features. It is to put the answer a buyer needs in rendered text rather than assume a model will extract it from markup. Pay once for clean templates, not every month for a “schema sprint.”

Line 5: Crawl access is worth fixing, not watching indefinitely

Here is the first line I would keep. PerplexityBot crawls and indexes pages for surfacing and linking, and respects robots.txt; Perplexity-User generally makes user-requested fetches and ignores it. Those are different behaviors, which is precisely why a generic “AI bots enabled” checkbox is not much of an audit.

The documented fix is small: Allow: / for search bots, IP allowlisting if a tight firewall is blocking them, and roughly 24 hours for the change to take effect. I have seen accounts where access was the whole blocker. If a bot cannot read a page, the copy, schema, and tracking around that page cannot rescue it.

Pay to check and fix crawl access. Do not pay a monthly premium to observe a rule that has already been corrected.

Line 6: Refresh the pages; do not mistake entity cleanup for citations

“Entity optimization” usually means consistent names, addresses, categories, and same-as links across a site and its profiles. It can help with disambiguation, and I would get the basics right. I would not call a tidy About page a citation strategy.

The content-refresh half has a stronger case. In a review of 7,387 citations, product pages accounted for 24.1%, articles 17.4%, comparisons around 13%, and homepages 7.8%. The draft’s cited research also found that half of citations came from articles published in the prior 11 months. Neither finding guarantees that updating a page will get it cited. Together, they give me a better place to spend the retainer than profile cleanup: pages that answer purchase questions with dates, prices, specs, and limits stated plainly.

Keep basic entity cleanup. Judge the ongoing fee by the buyer-question pages actually refreshed.

Line 7: Question-led pages are the core deliverable

Now the proposal reaches work I can inspect. I do not mean a generic service page with a new FAQ block. I mean a dated page that answers one buyer question with details a model can use: what it costs, who it fits, what breaks, or how long it takes. Each page should map back to a prompt from line 2, so the tracking has a job beyond producing a report.

I used to tell clients a strong service page was enough. I was wrong. A specific answer gives a model less to infer than a broad pitch does. It also gives a human reader a reason to stay. That is the mechanism I want the agency to work on, not a promise that publishing alone will produce citations.

If the retainer ships two to four such pages a month, each tied to a buyer prompt, I would keep this line. If it ships one 800-word trends post, I would not call that a content program. Count pages shipped and questions answered, not words billed.

Line 8: Third-party mentions deserve a real outreach plan

The final line is the other one I would keep. It is also easy to underweight because outreach takes more work than editing a page you control. In a study of more than a million AI-cited links, around 94% came from non-paid sources, including 82% from earned media. Across 75,000 brands, web mentions correlated at 0.664 with AI Overview visibility, versus 0.218 for backlinks. Correlation is not a placement guarantee. It is a reason to ask where the outreach hours go.

I want to see a plan for mentions on trade publications, relevant comparisons, and directories with real traffic. What the deck calls synergy, I call getting someone outside your company to talk about you. Editing your own site cannot do that job.

If the proposal has no earned-placement plan, its “citations” line is a label, not a plan.

What groas’s bot logs show—and what they do not

After those eight lines, I checked the traffic behind the argument. The figures below come from groas edge logs for groas-controlled pages. They count requests in live-answer traffic; they are not a cross-client benchmark or a count of confirmed citations.

The gap is striking. The two changelog posts received far more logged requests than the service page, and robots.txt received the most. But a request is not proof of a citation. The logs cannot tell me that each fetched post appeared in an answer, why the bot chose it, or whether the same pattern holds on your domain. Bot behavior can change, too.

They can tell me where to look first. Check access. Inspect what the frequently fetched, dated pages actually say. Then compare them with the generic page before commissioning another dashboard. That is a useful order of operations without pretending the table proves more than it does.

The shorter proposal I would sign

I would replace the eight-line retainer with three accountable workstreams:

1. Crawl access and technical fixes: audit the rules, fix blockers, and monitor for new ones.
2. Dated buyer-question pages: ship two to four a month, each tied to a tracked prompt.
3. Earned mentions: name the outreach targets and report links that actually went live.

That is close to how groas describes SEO & AI Search at $1,499 a month: new pages and technical fixes every day, work on third-party sources, logged actions, and a named strategist accountable for the result. It is a sharper alternative to paying $4,200 for a package whose monitoring lines can crowd out production. For more on the reporting problem, read the AI visibility report teardown. For the choice between pages you own and mentions elsewhere, read your blog vs third-party pages.

Four questions to ask before signing

On the sales call, I would ask for the work behind each answer, not another visibility adjective:

• Pages: How many dated buyer-question pages ship each month, and which prompt does each target?
• Mentions: What earned mentions went live last month? Show links, not pitches sent.
• Tracking: When share cited drops, which page gets written or fixed?
• Fees: Which charges are one-time setup, and which buy monthly work?

I would also ask what citation lift the agency has measured from llms.txt or schema before accepting either as a recurring charge. If the answers stay vague, the retainer is monitoring with a content garnish. You can compare the quote with this $3,000 AEO retainer teardown before signing.

My verdict on this proposal is to cut the score, the 50-prompt screenshot, the recurring llms.txt charge, and the monthly schema sprint. I would keep the access audit, the question-led pages, and the outside-mention plan, with modest prompt tracking to connect them. Pay for pages shipped and mentions earned, not scores watched. Anything else is the agency billing you to watch the game from the stands.