An AI visibility routine for a two-person marketing team

Why a routine beats a project

AI answers change every day. The same question can get a different answer on Tuesday than on Monday. A one-off audit is out of date in a fortnight. What a small team needs is not a bigger project but a smaller, repeating one: the same questions, the same measures, a short list of actions, a fixed slot in the calendar.

Set up once

Do these five things before the first weekly review. They take a few days of part-time work and almost never need to be redone.

A fixed prompt set

Choose 20 to 40 questions your buyers actually ask an assistant, across the stages of a purchase: category questions ("best scheduling tool for agencies"), comparison questions ("Fernwick vs Calendly for a 10-person team"), problem questions ("how do I stop double bookings with freelancers") and brand questions ("is Fernwick good for agencies"). Write them the way people type, not the way you would search Google. The full method is in how to choose the prompts to track. Then freeze the set. A prompt set that changes every week cannot show a trend.

Named competitors

List the five to eight companies that appear in answers to your prompts, not the ones on your sales battlecards. Run the prompt set once and note who gets named.

A fact sheet

One page, kept by one person, with the facts an assistant should get right: what the product does in one sentence, who it is for, pricing tiers and starting price, integrations, where the company is based, founding year, headcount band, and the three things it does not do. Keep it in a shared document and date each change.

Crawler access checked

Confirm that the AI crawlers can read your site: GPTBot, OAI-SearchBot, ClaudeBot, PerplexityBot and Google-Extended in robots.txt, no bot-blocking rules at the CDN or firewall, and the important pages rendering as plain HTML without JavaScript. Blocked crawlers are the most common reason a good site gets no citations, and the check takes ten minutes. AI crawlers explained lists each bot and what it is for.

A baseline across engines

Run the full prompt set against ChatGPT, Perplexity, Gemini and Claude, and check Google for AI Overviews on the same questions. Record for each prompt and engine: were you named, in what position, with what sentiment, which pages were cited. This is week zero, and the monthly report is read against it.

The weekly review: 60 to 90 minutes

Pick a fixed slot, the same morning each week, and keep to the time; the limit forces a choice.

1. Prompts you lost or gained (15 minutes)

Look only at prompts whose result changed since last week: named last week and not this week, or the reverse, or a move of more than a couple of positions. For each change, open the answer and read it. Most moves have a visible cause: a new page the engine found, a competitor's new comparison article, a review that was quoted.

2. The new citations (15 minutes)

List the pages cited this week that were not cited last week, across all prompts and engines, for you and for competitors. Sort into three piles: your own pages (good, note which ones), third-party pages that mention you (check what they say), and third-party pages that mention competitors but not you (the gap). That third pile is the richest source of actions a small team has.

3. Pick at most three actions (20 minutes)

From what you saw, choose up to three actions. There are only three kinds:

• Fix a page: a wrong fact, a missing sentence, a page that does not say who it is for, a pricing page that hides the starting price.
• Write a page: a comparison, a use-case page, a plain answer to a question the engines keep answering with someone else's page.
• Get onto a third-party page: a review, a listicle, a forum thread, a partner's integrations directory.

Write each one as a sentence with an owner and a date: "Rewrite the integrations page to name all 14 integrations in plain text. Owner: Dana. Due: Friday." Put them on the same list as the rest of the team's work. In Echo, this list is the Actions page and each action keeps its proof and a before-and-after; a shared document works too.

4. Ship one (30 minutes)

Do the smallest of the three now, in the remaining time. A fact fixed on a page, a review requested from two customers, a paragraph added to a comparison page. Shipping inside the slot guarantees progress: a team that reviews and never ships has a dashboard, not a routine.

The monthly report: one page

On the first review of the month, replace the weekly session with a report. It should fit on one page and take an hour to assemble if the weekly notes exist.

1. Share of voice per engine. For each engine, your share of all brand mentions in the answers against each named competitor, this month and last. Share of voice in AI answers covers the calculation. Report the engines separately; an average hides that Gemini and ChatGPT often disagree.
2. The pages that got cited. Your ten most-cited pages this month, with how many answers cited each. Note which were written or fixed as actions.
3. AI referral visits and what they did. From GA4, the sessions from chatgpt.com, perplexity.ai, gemini.google.com and the rest, the landing pages, and the conversions. How to see this in GA4 walks through the set-up.
4. Wrong facts found and fixed. The list from the fact sheet: each wrong fact, where it appeared, whether it is fixed on your site, whether the third-party source is fixed, whether the answers now have it right.
5. Three decisions for next month. Not a list of ideas. Three choices with names attached: which page to write, which objection to answer with proof, which third-party source to pursue.

Keep every month's report in the same folder; by the third month the trend is the story.

The quarterly review: two hours

Every three months, step back and review the set-up itself.

• The prompt set. Drop prompts that no buyer would ask or that no engine ever answers with a brand. Add prompts from sales calls, support tickets and the questions you see in Search Console. Change at most a quarter of the set, so the trend survives.
• The competitor list. Remove the ones who stopped appearing in answers. Add any new name that appeared in more than a few answers this quarter.
• The fact sheet. Reread every line against the product. Prices, tiers, integrations and headcount change, and a stale fact sheet produces confident wrong fixes.
• Which content formats worked. Look at the pages that earned citations and group them: comparison pages, use-case pages, plain how-to answers, glossary pages, pricing pages, customer stories. Write more of the format that got cited and less of the one that did not.

What to skip

A short routine survives because of what it leaves out.

Daily checking. Answers vary from run to run, and a daily look at one prompt produces anxiety and no information. If a tool runs the prompts daily, good; read the results weekly.

Chasing one answer. A single answer that leaves you out, or calls you expensive, is a data point. Three weeks of the same answer on the same prompt is a pattern, and a pattern has a source you can fix.

Rewriting everything for AI. There is no secret formatting that makes engines cite a page. The pages that get cited answer a question plainly, state facts in sentences, name who the product is for, and are linked from elsewhere. Fix the pages your prompts lead to, in order of impact, and leave the rest alone.

How to split the work between two people

The split that works is by kind of work, not by engine or by week.

• Person A owns the data and the facts: runs or reads the prompt results, keeps the fact sheet, assembles the monthly report, checks crawler access after every site deploy, and owns the "fix a page" actions.
• Person B owns the content and the outreach: writes the new pages, requests reviews and customer stories, pitches the listicles and directories, replies in threads, and owns the "write a page" and "get onto a third-party page" actions.

Both attend the weekly review; A runs it, B keeps the action list. The monthly report is A's, the three decisions are made together. If one of the two is also running paid, email or events, the AI slot stays at 90 minutes and the actions shrink to two; the rhythm matters more than the volume.

The one-page template

Copy this into a shared document.

Weekly review (date, 90 minutes)

• Prompts lost since last week (prompt, engine, what the answer now says)
• Prompts gained since last week (prompt, engine, which page was cited)
• New citations of our pages (page, prompts, engine)
• New citations of competitors only (page, who is named, why)
• Actions chosen (at most three: what, owner, due date, kind: fix, write, get onto)
• Shipped today (one item, link)

Monthly report (month, one page)

• Share of voice per engine vs each competitor, this month and last
• Our ten most-cited pages, with answer counts, and which were actions
• AI referral sessions by source, top landing pages, conversions
• Wrong facts found, fixed on site, fixed at source, now correct in answers
• Actions shipped this month and their before-and-after
• Three decisions for next month (what, owner)

Quarterly review (quarter, two hours)

• Prompts to drop and add (at most a quarter of the set)
• Competitors to drop and add
• Fact sheet changes
• Formats that earned citations, formats that did not
• One change to the routine itself