What AI SEO Services Actually Include, and What to Refuse to Pay For

A real AI SEO scope is about eleven line items, and on our own measurement only some of them change whether an engine cites you. Three get billed with no evidence behind them at all: an llms.txt file, “proprietary AEO schema”, and a single blended AI visibility score. Refuse those three, and buy the rest by named deliverable.

At a glance

• The scope, in one breath: entity and coverage mapping, answer-first page structure, schema and technical hygiene, internal linking, content production, refresh, off-site placement, per-engine citation polling.
• What our own polling shows: on LeadsNow’s fixed prompt registry, 1–13 September 2026, leadsnow.ai was cited at least once for 5 of 30 prompts on ChatGPT (browsing) and 10 of 35 on Gemini (grounded), against 74 of 126 prompts that surfaced a leadsnow.ai page on DuckDuckGo.
• The three to strike from a proposal: llms.txt as a deliverable, AI-specific markup, and any single blended “AI visibility score”.
• The test: every line must name an artefact you receive, a number that would change if it worked, and the engine and window that number is measured on.

What do AI SEO services actually include?

Strip the naming and every credible AI SEO service, AI SEO agency and in-house AI SEO specialist does the same eleven jobs. The scope does not change with the label on the invoice, only who does it and how much gets done.

1. Entity and coverage mapping — who the engines think you are, and which buyer questions you have no page for.
2. Answer-first page structure — the answer in the opening paragraph, question-shaped headings, real HTML tables.
3. Technical crawl, indexation and schema validity.
4. Internal linking in both directions — new page into its hub, hub back into the page.
5. Content production against the coverage map, for questions that were counted rather than brainstormed.
6. Refresh on a named cadence — the shape of it is in our note on how often to update content for AI search.
7. Off-site placement — being named on domains you do not own.
8. Per-engine citation polling against a fixed prompt set.
9. Reporting split by engine, with the raw per-prompt log attached.
10. A named owner for truth — your case studies, your numbers, your product reality. No vendor supplies this.
11. Whatever the vendor calls the AI-specific part. This is where the padding lives.

An AI SEO service scope is eleven jobs long, and the eleventh is the one to read carefully.

Which line items measurably move citation share, and which are just billed

Verdicts drawn from LeadsNow’s own citation polling and from sources opened at the primary. Where there is no evidence either way, the row says so. AirOps’ 2026 State of AI Search is a vendor publication that does not disclose its engines, prompt count, sample size or date range, so it is named in every cell using it.

Row seven is uncomfortable for us: LeadsNow sells on-site work, and the only public figure we could find — AirOps’, on a methodology it does not disclose — puts most brand mentions inside AI answers on domains we do not touch.

What our own polling says about a single “AI visibility score”

LeadsNow runs an AEO monitor against its own domain first: a fixed registry of buyer-phrased prompts is asked of ChatGPT with browsing and Gemini with grounding on a schedule, and each answer parsed for whether leadsnow.ai is cited. Between 1 and 13 September 2026, on prompts registered before 1 September, leadsnow.ai was cited at least once for 5 of 30 prompts on ChatGPT (17%) and 10 of 35 on Gemini (29%) — while the same registry run through DuckDuckGo surfaced a leadsnow.ai page for 74 of 126 prompts (59%). The unit is prompts, not individual answers: each prompt was polled several times in the fortnight.

That is one site, one prompt set, two engines and a fortnight — not an industry benchmark. It still settles a purchasing question. Averaged into one score those become a number true of no engine, and the 59% DuckDuckGo figure drags the blend upwards while saying nothing about whether an assistant quotes you. Ask for share of answer reported per engine, with the denominator printed beside it.

Is an “AI SEO specialist” a different scope from an AI SEO agency?

No — the eleven line items are identical. People search for ai seo specialists and ai seo agency as if they were different products; they are the same product bought in different units. One salaried person can own the mapping, the technical work and the truth, then runs out of hours around content and polling. A team carries all eleven and will never own line ten, your proof and your product reality.

The failure mode to price in is that nobody owns line ten. Our breakdown of what an AI SEO retainer delivers week by week sets out how the cadence divides.

What the scope costs in hours if you run it in-house

Price the same work as your own time, substituting your inputs:

• Measurement. 40 prompts × 2 engines = 80 polls a cycle, fortnightly = 160 a month. By hand — ask, read the citations, log the domains — budget three minutes each: 8 hours a month.
• Publishing. 8 pages a month at 4 hours each, including opening every external number at source: 32 hours.
• Refresh. A 100-page library, half of it re-touched within six months, is roughly 8 pages a month at an hour each: 8 hours.
• Total: about 48 hours a month — a quarter of a full-time person, before any technical or off-site work.

That arithmetic is the buying decision, yours to do rather than ours. Where the crossover sits:

Measurement is the line item that gets silently dropped in-house, and it is the only one that tells you whether the other ten worked.

The Named-Deliverable Test: how to read a scope of work in ten minutes

Give every line in the proposal three columns. A line that cannot fill all three is a category, not a deliverable.

1. The artefact. What lands in your inbox or CMS because this line exists? “A refresh log with URL, date and what changed” passes; “ongoing AEO strategy” does not.
2. The number. Which figure differs in 90 days if this line worked, and what is it today? If nobody can state today’s value, the line cannot be assessed later.
3. The engine and window. Measured on which engine, over what period, against how many prompts? A citation rate with no denominator is not a measurement.

Run it on the eleven items and the padding falls out. llms.txt fails column two: no number moves. Proprietary AEO markup fails column three: Google says there is nothing there to measure. A blended score fails column three too, having no single engine behind it. The long version of the first is our write-up of what the llms.txt server-log evidence actually shows.

A scope that concedes in writing that this month’s pages will not be cited this month, that on the only published figure most brand mentions sit on domains you do not own, and that no citation can be guaranteed, is one you can hold someone to.

If a line in an AI SEO proposal cannot name an artefact, a number and an engine, you are being sold a category.

Frequently asked questions

Should an AI SEO proposal include an llms.txt file as a paid deliverable?

No. Ahrefs studied all 137,210 domains in its Web Analytics population that received traffic in May 2026 and found that of the roughly 38,000 serving a valid llms.txt file, 97% saw no requests for it whatsoever that month; AI retrieval bots made 1.1% of the requests those files did receive, 233 fetches in total. Publish it as a five-minute chore if you like. Paying for it, or accepting it as evidence a vendor is doing AEO, is not defensible.

Is “AI SEO specialist” a real role or a renamed SEO job?

A real role with one added deliverable, not a new discipline. The scope is the same eleven line items SEO always had, plus one that did not exist before: polling a fixed prompt set against named engines and logging which sources each answer cites. If a specialist’s scope has no measurement line, you have hired a renamed SEO. If it has one, ask which engines and how many prompts — those two numbers decide whether the reporting means anything.

Can an AI SEO agency guarantee citations in ChatGPT or AI Overviews?

No, and a guarantee is a reason to stop reading. Google’s documentation on AI features in Google Search states that “there are no additional requirements to appear in AI Overviews or AI Mode, nor other special optimizations necessary” and that there is “no special schema.org structured data that you need to add”. No documented lever, no guarantee. What can be committed to is a fixed prompt set, a baseline, and a per-engine rate on a schedule.

How do I check whether an agency’s AI search reporting is real?

Ask for the raw per-prompt log for one past month: prompt text, engine, date, and the domains cited in each answer. Real monitoring produces a boring table with plenty of rows where you do not appear. A screenshot of one favourable ChatGPT answer is not evidence: the same prompt asked twice can return different sources. Check the denominator has not moved between reports — a rate that improves because the prompt set shrank is not an improvement.

What should I refuse to pay for in an AI SEO scope of work?

Three things: an llms.txt file as a priced deliverable, anything sold as proprietary AEO schema or AI-specific markup, and a single blended AI visibility score in place of per-engine rates. Beyond those, refuse any line failing the Named-Deliverable Test: if it cannot name the artefact you receive, the number that would change, and the engine and window it is measured on, it is a category heading with a price.