Skip to content

What should an AI visibility retainer deliver each month? A checklist before you sign

Split an AI visibility retainer into monitoring, on-site fixes and off-site work, then compare proposals line by line. The red flags to check before you sign.

By Published Updated 7 min read
Costs and buying, 7 min read — Three floating glass shelves, each holding a row of differently shaped glowing objects, crossed by a single thin beam of light.

The short answer

Each month a retainer should deliver three things you can check: measurement on a named prompt list across named engines, on-site fixes with before-and-after evidence, and off-site work such as earned mentions and corrected listings. Ask for each as a separate line. A proposal that cannot list its prompts or explain its score is not ready to sign.

Key takeaways

  • In AgencyAnalytics' survey of 494 agency professionals (February–April 2026), 66% reported increased client demand for AI search services.
  • G2 listed more than 150 answer engine optimisation products by January 2026, up from 7 in March 2025, so buyers face many similar-sounding offers.
  • Split every retainer into monitoring, on-site work and off-site work, and price each line separately so proposals can be compared.
  • The red flags are an unexplained score, no prompt list, no named engines, rankings presented as AI results and guaranteed placements.
  • Our view: ask for the prompt list and one sample monthly report before you sign anything.

Why retainers are hard to compare

AI visibility retainers are hard to compare because each provider bundles different work under the same name. One sells a dashboard with a monthly call. Another sells content. A third sells outreach and calls it generative engine optimisation. All three may quote a similar monthly fee.

Demand is rising faster than the vocabulary has settled. In AgencyAnalytics' 2026 benchmarks survey of 494 agency professionals, 64% of them in the US, fielded February–April 2026, 66% reported increased client demand for AI search services, and 64% named Google's AI Overviews as their top concern.1 AgencyAnalytics sells reporting software to agencies.

The software side has grown as quickly. G2 listed more than 150 products in its answer engine optimisation category by January 2026, up from 7 in March 2025.2 Larger platforms are buying in too: Sitecore announced in June 2026 that it had acquired the AI visibility platform Scrunch.3 A buyer comparing proposals is choosing from a crowded, young market.

The three parts of a retainer

A complete retainer has three parts: monitoring, on-site work and off-site work. Ask each provider to price them as separate lines, even if they sell them together. That one request makes proposals comparable.

Three lines every proposal should separate

Monitoring

  • Named prompt list, refreshed quarterly
  • Named engines and countries
  • Repeated runs, with a margin of error
  • Competitor comparison

On-site work

  • Pages rewritten to answer first
  • Crawler access and structured data checks
  • Entity facts kept consistent
  • Before-and-after evidence per change

Off-site work

  • Mentions earned on sites engines cite
  • Directory and review listings corrected
  • Wrong facts traced and fixed at source
  • A log of every placement attempted
Illustrative. Our framing of a complete retainer; off-site work is the line most often missing.

Monitoring

Monitoring tells you whether anything changed. It should name the prompts, the engines and the countries, and run each prompt more than once, because AI answers vary from run to run. Our measurement spec for AI visibility scores sets out what to ask for, including the margin of error.

On-site work

On-site work changes your own pages: answer-first rewrites, crawler access, structured data and consistent facts about the company. Each change should come with the prompts it was meant to affect, so the next month's measurement can show whether it did.

Off-site work

Off-site work changes what others say about you: earned mentions on the domains engines already cite, corrected directory and review listings, and fixes to wrong facts at their source. It is slower and harder to fake. Our view: a retainer with no off-site line is a content or reporting contract, whatever it is called.

What should arrive each month?

Each month should bring a short report you could hand to a colleague without explanation. The table below is the checklist we would use to judge one.

Monthly deliverableWhat good looks like
Share of answer by engineWith a margin of error and last month's figure
Prompt listAttached, with any changes marked and explained
Citation sourcesWhich domains engines cited for your prompts
Errors foundWrong facts, with engine, date and source traced
On-site changes shippedPage, date, change and the prompts it targets
Off-site work doneMentions earned, listings fixed, attempts that failed
Next month's planThree to five actions, each tied to a prompt group

Our checklist. A monthly report missing two or more rows is a reason to ask questions.

Notice what is not on the list: rankings, traffic totals and the number of articles published. They can sit in an appendix. They are not evidence of AI visibility on their own, and a report that leads with them is answering a different question. If your board wants a short version, our guide to which AI search numbers belong in a board report sets out four lines to carry upward.

Red flags in a proposal

Five red flags are worth checking before you sign, and each can be tested with one question.

  • An unexplained score. If the provider cannot say how the score is calculated, from which prompts and engines, it cannot tell you why it moved. Ask for the formula.
  • No prompt list. Without the list, nobody can check that the prompts match what your buyers ask. Ask to see it before signing.
  • No named engines. "AI search" is not a measurement. Ask which engines, in which countries, and how each is queried.
  • Rankings dressed as AI results. A keyword ranking report with a new title is still a ranking report. Ask where the AI answers themselves appear.
  • Guaranteed placements. No provider controls what an engine says. A promise of a named position in ChatGPT is a promise nobody can keep.

Marketers are already asking harder questions. Digiday reported in May 2026 that marketers were questioning the price of AI visibility tools as results varied between runs and between tools.4 That scepticism is healthy, and the answer to it is a method you can inspect, not a lower price.

A note on our own terms, because this list could be read against us. For qualified programmes, Sigzen AI commits to keep working at no charge if you are not cited within 90 days. That is a commitment about our work, decided after an audit and stated in writing. It is not a promise of a position, and we would treat any guarantee of a specific ranking or placement as a red flag, including from us.

How to compare two proposals line by line

Put the proposals in one table with the three lines as rows, then fill in what each provider will deliver and at what price. Gaps show up at once.

Neither proposal is wrong. They buy different things, and the table makes the trade visible. Northwind can then ask Acme to price an off-site line, or ask Contoso to add engines, rather than choosing on the headline fee. Our dated list of published GEO prices shows what agencies and tools charge for each kind of work.

Our view: price per line matters more than the total, because the off-site line is where most of the hours go and where cheap retainers cut first.

Where our programme sits

A disclosure: we sell a retainer. Our growth programme is $3,500 a month on a six-month minimum, with part of the fee linked to share of answer, as listed on our pricing page. It includes monthly prompt-set runs on six engines (ChatGPT, Gemini, Perplexity, Claude, Copilot and Google AI) through our AI answer monitor, 8–20 human-edited, answer-shaped assets a month, directory, review and editorial work, and a weekly dashboard with a quarterly prompt-set refresh.

Hold us to the same checklist. Ask for the prompt list, a sample monthly report and the off-site log. If the programme is not the right fit, the audit on our AI visibility page gives you a baseline you can take to any provider.

Sources

  1. AgencyAnalytics, 2026 Marketing Agency Benchmarks Report: 494 agency professionals, 64% in the US, Feb–Apr 2026 (Jun 2026)
  2. G2, growth of the AEO software category: products listed on G2, Mar 2025–Jan 2026 (Jan 2026)
  3. Sitecore, Sitecore acquires Scrunch to help brands influence discovery and buying decisions (Jun 2026)
  4. Digiday, Marketers question expensive AI visibility tools as inconsistent results fuel skepticism (May 2026)

Questions readers ask

  • Long enough for off-site work to take effect, which is usually a quarter or more. Agree a baseline and a written expectation at the start, then judge the first quarter against them. Monitoring and on-site changes should show progress sooner, so ask for monthly evidence of what shipped even before share of answer moves.

Keep reading

Free, in two minutes. Enter your domain and we'll score it against three competitors across six engines.

No account needed. The report is emailed within 24 hours.