What answer engine optimization services should deliver

Whoever you hire, the output should be inspectable. Ask for these by name:

• A question inventory. The real questions your buyers ask, pulled from Search Console queries, People Also Ask boxes and keyword data, not brainstormed in a workshop.
• One page per question, answer first. The opening paragraph answers the question in plain terms; the rest of the page earns the right to be believed.
• A quotable definition on every page. One or two sentences that stand alone when lifted out of context.
• FAQ blocks with FAQPage JSON-LD. Follow-up questions answered in a short paragraph each, marked up so the answers are machine-readable.
• Crawler access. Confirmation that the fetchers can reach the pages, checked per user agent: OAI-SearchBot for ChatGPT search, PerplexityBot for Perplexity, GPTBot for OpenAI training, Googlebot for everything Google. Google-Extended is not a fetcher and does not belong in that list: it governs whether pages Googlebot already has are used to train Gemini, and blocking it does not remove you from AI Overviews.
• First-party facts. Numbers and specifics from your own product data, datasheets or support history that no competitor's page can carry.
• A citation report. Which questions were put to which assistant, and whether your site was named.

If a proposal lists "AEO strategy" and "AI readiness" without any of these nouns attached, you are being sold a retainer, not a result. Every item above is a file, a page or a report you can open and check.

Most AEO contracts fail at the acceptance criteria rather than the effort. Each line below pairs a deliverable with the thing you check to know it happened.

• A published question set. Deliverable: the exact questions, written down before work starts. Check: you can read the list, it does not change between reports, and changing it starts a new series rather than being spliced into the old one.
• Answer-first pages. Deliverable: one question per H2, answered in the first forty to sixty words. Check: read the first paragraph under any heading on its own. If it needs the paragraph above it to make sense, it will not survive being chunked by a retrieval layer.
• A checkable fact per claim. Deliverable: a number, date, specification or named source behind anything a reader could dispute. Check: pick three claims at random and try to verify them.
• Schema that matches the page. Deliverable: Article and FAQPage JSON-LD generated from the same content the reader sees. Check: run the URL through Google's Rich Results Test and compare the extracted questions with the visible ones.
• Verified crawler access. Deliverable: confirmation that the fetchers you care about are allowed. Check: fetch /robots.txt yourself and read the group named for that crawler, not the wildcard group — a crawler obeys only its most specific matching group and ignores the rest, so a rule in the wildcard group never reaches a bot with a group of its own.
• An indexed-share figure. Deliverable: what percentage of pages published for you Google has actually indexed. Check: Search Console's page indexing report. This is the number that says whether publishing became visibility, and it is the one most reports omit.
• Probes with retrieval on. Deliverable: assistant checks that use web search, not the bare model. Check: ask whether the answers carried links. Ours did not for five weeks, which is how we found out our own probe was measuring brand fame instead of our pages.
• A takedown route. Deliverable: how a live page comes down, and how fast. Check: do it once, in the first month, on a page you do not need.

Answer engine optimization services are sold three ways, and the billing model tells you what you are really paying for.

Agency retainer. A monthly fee buys a team's time: an audit, a content plan, a fixed number of new or rewritten pages, and a report. You pay whether or not the pages get cited, and output is capped at what people can write and review in a month. Good agencies bring judgement, and you pay for it by the hour.

Software subscription. A monthly licence to a tool that drafts or scores content. Cheaper per page, but briefing, editing, publishing and measuring stay with you, and the bill is the same whether you publish one page or fifty.

Per published article. You pay for finished pages that go live, and nothing for drafts that fail review. That is IT Master's model: a prepaid balance, no subscription, four quality tiers, and a price per published article set out on the pricing page.

None of these is wrong. A retainer suits a business that needs strategy more than pages. Software suits a team with writers and time. Per-article suits anyone who wants checked, published pages at a steady cadence without hiring for it.

You can learn most of what matters without a contract. The whole exercise takes a few hours spread over a week.

• Day one — set your own baseline. Write down twenty questions your buyers ask, in their words. Put each to ChatGPT with search on and to Perplexity, and record which domains are cited. Keep the file. Whatever any vendor later reports, you have an independent starting point, and it cost you nothing.
• Day one — check the door. Fetch your own /robots.txt and read the groups for OAI-SearchBot, ChatGPT-User, PerplexityBot and Googlebot individually. Then view the served HTML of one important page and confirm the body copy is in it rather than injected after load. Our free AI crawler check does both in a few seconds.
• Day two — ask for one page they published last month. Not a portfolio, one live URL. Read the first paragraph under each heading in isolation. Check the JSON-LD against the visible text. Look for a named author and a date. Click every outbound link.
• Day three — ask what they refuse to promise. A supplier who will not name a limit has not thought about one.
• Day four — ask for their reject rate. Anyone producing at volume knows it. Ours is published on the AI Content Index as a first-pass rate per tier, computed from the ledger, including the tiers where it is poor.
• Day five — price a failure. What do you pay for a draft their own checks reject? If the answer is the full rate, nothing in the system has an incentive to reject anything.

If a provider passes that week, the contract is mostly a formality. If they cannot produce a live URL and a reject rate, the contract will not save you.

Each article runs through one pipeline, with the AEO deliverables built in rather than bolted on.

Discovery pulls real demand from DataForSEO and your Search Console, so every topic starts as a question someone typed. Research reads the competing pages, your first-party data (product records, datasheets, support Q&A) and relevant forum threads, then finds what the competitors left out.

A Claude model writes a question-led article: the direct answer in the opening, a quotable definition, sections that follow the buyer's follow-up questions, and an FAQ block. The draft then enters a validation gauntlet: a fact-check and a novelty score against the competitors by Gemini and GPT judges, plus an E-E-A-T critique and AI-tell detection, none of it by the model that wrote the draft, with up to three revision rounds. Across 186 Standard-tier runs, 21% of drafts cleared every check on the first attempt with no revision, and 51% got there within the three revisions allowed; the rest never published. Pro-tier judges are stricter: 25% cleared every check across a small sample of 16 runs.

A passing article publishes to your site with JSON-LD schema including the FAQ, internal links into your existing pages, a hero image and an IndexNow ping to the search engines. The full sequence is on the how it works page.

The point of AEO is to be named as a source, so the measurement has to ask the assistants directly. Every night IT Master puts a fixed set of your buyers' questions to GPT and Perplexity and records whether your site is cited, alongside the Search Console clicks and impressions for the same pages. Having both numbers is how you tell a page that ranks from a page that gets quoted.

The same nightly loop watches for decay. When a page's impressions slide or its facts go stale, it is queued for a refresh rather than left to sink.

Two things this measurement will not do. It will not promise a citation: assistants change their sources week to week, and any vendor who promises you placement is guessing. And a week of data is noise; judge the trend over a quarter.

You can run a version of these checks yourself first. The AI visibility checker puts buyer questions to the assistants and shows whether your pages appear as sources, the AEO checker scores any URL on nine deterministic checks, and the AI crawler check reads your robots.txt to show which assistant crawlers you allow or block.

Crawling, indexing, ranking and citation each move on their own schedule, and each gates the next. Watch them in order.

• Weeks 1–2: baseline and access. The question set is fixed and probed once, misses recorded. Crawler access is verified per user agent. The indexed share of your existing pages is measured. Nothing published yet is worth judging.
• Weeks 2–6: publication and discovery. Pages go live and are submitted through IndexNow, with fresh sitemap lastmod and internal links from pages that are already crawled often. The signal to watch is not rankings; it is whether Search Console moves the new URLs from unknown to crawled to indexed.
• Weeks 4–10: impressions. Pages start appearing for the questions. This is the first honest evidence that the topic and the phrasing were right.
• Weeks 6–12+: extraction and citation. Featured snippets and People Also Ask entries come from first-page results, so they follow position. Assistant citations can come earlier, because retrieval works on narrower sub-queries than the head term, and Perplexity is usually where a new site appears first.

Where programmes stall, in our experience, is stage two rather than stage four. On four sites we publish for, measured on 25 August 2026, the share of published articles Google had indexed ran from 25 of 27 down to 12 of 25 — same pipeline, same week. On the worst of them, most misses were reported as "URL is unknown to Google" with no crawl recorded: never discovered, not judged and found wanting. If your provider is discussing passage structure while your indexed share is under 80%, they are optimising the wrong stage.

Nothing here requires a new CMS. On WordPress, install the plugin, paste one key, and articles arrive as posts with the Yoast or Rank Math fields filled. On Next.js, @itmaster/sdk renders the feed. Anything else can consume the REST pull feed or receive a signed push webhook at any endpoint. Shopify and Webflow we set up for you on request; Wix and Squarespace get a copy-and-paste queue.

The engine runs today on live UK CCTV trade and retail sites. Each is grounded in whatever first-party data that site holds: product records, datasheets, support questions. Editorial responsibility sits with Editor-in-Chief Rodd Azad, a software engineer and the founder of IT Master and code-nest.dev.

If you want to know where you stand before spending anything, run the free site check. It shows what search engines and AI assistants currently find on your website, and what we would write first. No account, no card.

Primary sources for the claims on this page, all read on 4 September 2026.

• Google Search Central, "Robots.txt specification" — a crawler obeys only the most specific matching group and ignores every other group: developers.google.com/search/docs/crawling-indexing/robots/robots_txt
• Google Search Central, "AI features and your website" — AI Overviews and AI Mode use the ordinary Googlebot crawl and need no separate optimisation: developers.google.com/search/docs/appearance/ai-features
• Google Search Central, documentation changelog — the FAQ rich result stopped appearing on 7 May 2026 and the FAQPage documentation was removed in June 2026, so treat FAQPage markup as machine-readable structure rather than a search feature: developers.google.com/search/updates
• Google's Rich Results Test, for checking that your schema matches your visible page: search.google.com/test/rich-results
• OpenAI, "Bots" — OAI-SearchBot, ChatGPT-User and GPTBot: developers.openai.com/api/docs/bots
• Perplexity, "Bots" — PerplexityBot and Perplexity-User: docs.perplexity.ai/guides/bots
• IndexNow protocol documentation: www.indexnow.org/documentation
• Our own figures, published rather than asserted: the AI Content Index and the weekly AI Citation Index.