How should an agency scope, price and report AI search as a service?

Why are brands asking for AI search now?

The request rarely arrives as a brief. It arrives as a question at the end of a monthly review: "Are we showing up in ChatGPT?" A founder has seen a rival named in an AI answer, or a customer has said they found the brand through one, and now it is on the agenda.

There are concrete reasons it is coming up. Google's AI Overviews and AI Mode now sit in Search results, and Search Console has a dedicated Search generative AI performance report that, as of 31 August 2026, is available to all websites. ChatGPT shows product results, and OpenAI's publishers and developers FAQ says it tags referral links with utm_source=chatgpt.com, so the visits are visible in GA4. Brands can now see that this traffic exists, which means they expect their agency to have a view on it.

The agency that already owns search, content and the website is the natural owner. The question is whether it sells a real, scoped service or a vague promise.

What does it cost to sell it badly?

Saying no cedes the work. If the agency has no answer, the brand will find another vendor, and that vendor may soon be asking about the rest of the retainer too.

Overpromising costs more. AI answers change by phrasing, location, user history and the day. An agency that promises "top placement in ChatGPT" will lose that argument within a month, because the brand can run a prompt and get a different answer. Google is explicit that "indexing and serving isn't guaranteed" in its guidance on AI features, and its generative AI optimisation guide says no third-party tool has access to its internal ranking or AI systems. Sell the work, not the placement.

What does an AI search service include?

Seven components cover the documented levers. The D2C guide to AI search explains the method behind each one; here they are as deliverables.

1. Baseline audit

Crawler access for Googlebot, OAI-SearchBot and PerplexityBot; brand facts across the site, Google Business Profile, marketplaces and social; structured data status; Search Console's Product snippets, Merchant listings and generative AI reports; GA4 referrals from chatgpt.com and perplexity.ai. The output is a dated snapshot you will compare against every month.

2. A prompt set

A fixed list of buying questions the brand's customers really ask, agreed with the brand and run in ChatGPT, Perplexity and Google on a schedule. It is the backbone of reporting, so keep it stable and version any changes.

3. Entity and schema fixes

A signed-off brand fact sheet, every surface corrected to match it, Organization markup with sameAs links to official profiles, and Product markup with accurate Offers on every buyable page.

4. Content that answers buying questions

One strong, first-hand page per real buying question: choosing, comparing, trusting, using. Not mass-produced variants, which Google's guide warns can breach its scaled content abuse policy.

5. Product data

Prices, stock, shipping and return policies kept consistent on the page, in the markup and in the Merchant Center feed. Flagged items in Search Console are worked down every month.

6. Mentions and reviews

Post-delivery review requests on Google Business Profile and marketplaces, replies to every review, genuine press and creator coverage, and correction of wrong facts wherever they appear. No bought mentions.

7. Monthly tracking

The prompt set, the generative AI report, GA4 referrals and the work log, reported in the same structure every month.

What should the agency never promise?

Put these in the statement of work as exclusions, in plain words.

• Guaranteed placement, rank or mention in any AI answer.
• A fixed number of AI citations or referrals by a date.
• That a single file, such as llms.txt, will change visibility. Google says Google Search does not use it.
• A "Google AI score" from a third-party tool presented as Google's view.
• Control over how a platform words its answer about the brand.

How do you package it into tiers?

Tier by how much the agency does, not by promised outcomes. The table below is an illustrative structure, not a market standard.

Swipe the table sideways to see every column.

Two pricing rules hold whatever the numbers. Price the audit separately, so the brand pays for the diagnosis even if it does not buy the retainer. And keep content volume as the main lever between tiers, because it is the part that costs the agency the most time.

What goes in the statement of work?

A short, specific outline protects both sides.

1. Objective. Improve how accurately and how often the brand is represented in AI search answers for an agreed prompt set, measured monthly.
2. Scope. The components in the chosen tier, with quantities.
3. Prompt set. The agreed prompts as an appendix, with how changes are proposed and approved.
4. Access needed. Search Console, GA4, the site CMS, Merchant Center, Google Business Profile and marketplace seller accounts.
5. Brand responsibilities. Fact sheet sign-off, approval turnaround for content and replies, product information.
6. Measures. Share of prompts naming the brand, generative AI impressions, GA4 referrals from AI sources, work shipped.
7. Exclusions. Everything in the "never promise" list above.
8. Reporting cadence and format. Monthly report, quarterly review.
9. Term and review point. A minimum term long enough for content and fixes to be crawled, with a review before renewal.

What does a monthly report look like?

Keep the same sections every month so the brand learns to read it.

1. Summary. Three lines: what moved, what was shipped, what is next.
2. Prompt set results. Share of prompts naming the brand on each platform, this month against last month and the baseline, with the prompts that changed listed.
3. Google AI results. Impressions from the Search generative AI performance report, and the pages most used as sources.
4. AI referrals. GA4 visits and new users from chatgpt.com and perplexity.ai.
5. Product data. Valid and flagged counts in Product snippets and Merchant listings.
6. Reviews and mentions. New reviews, replies sent, notable mentions, wrong facts found and fixed.
7. Work shipped. Every fix and page, with dates.
8. Next month. The planned work, and what the agency needs from the brand.

How do you show progress honestly when answers change daily?

Fix the prompt set. A changing set makes every trend meaningless. Add new prompts as a separate group and keep the core set intact.

Run each prompt more than once. Record whether the brand appeared in each run and report the share, not a single snapshot. A brand named in some runs and not others is a finding, not a failure.

Report trends, not days. Compare month on month and against the baseline. One day's answers prove nothing either way.

Separate inputs from outcomes. The agency controls what it ships; it does not control the answer. Report both side by side, so a flat month with strong delivery reads as what it is.

Use first-party data where you can. Search Console and GA4 are the brand's own records. Note their limits: the generative AI report shows impressions without queries or clicks, and not every AI answer produces a click.

Say what you do not know. If a number moved and you cannot tie it to a cause, say so. An agency that admits uncertainty keeps its credibility for the months it does have a win.

How do Moolwan's numbers make the case?

Moolwan, the Bengaluru home décor brand run by the ai-dm founders, is where this work is tested every day.

Use numbers like these the way you would in a pitch: as evidence that AI search is a real, measurable channel for a D2C brand, not as a promise that another brand will get the same result. They show where impressions and visits came from; they do not show which questions triggered them.

How do you sell it to existing brands first?

Start with the brands you already serve. You have access to their Search Console and GA4, you know their products, and they already trust your reporting.

1. Run the baseline quietly. Check crawler access, facts, schema, the generative AI report and GA4 AI referrals for each brand.
2. Bring one finding to the next review. A blocked crawler, a wrong price in a marketplace listing, a flagged product count. Something specific and fixable.
3. Offer the audit as a paid step. Diagnosis first, retainer second.
4. Propose the tier that fits. Start most brands at Foundation and move up when content becomes the bottleneck.
5. Report the first month in full. The first report is the sales document for the renewal.

For prospects, the same logic applies: lead with a specific finding, not a pitch.

How does ai-dm support the service?

ai-dm's AI search visibility tracking runs each brand's prompts across ChatGPT, Perplexity and Google AI results, so the monthly numbers come from a schedule, not screenshots. White-label reports carry the agency's branding and can be scheduled. The AI search services page shows how agencies package it, and the free agency AI visibility audit lets you see the method in action before you commit. ai-dm drafts content, schema and replies; the team approves before anything goes live.

What should you do next?

Pick two existing brands, run the baseline and bring one finding to each next review.

Related pages

• Prospect audit: win the first call with evidence

  A prospect audit checks a brand's search, AI answers, ads, site and reviews from its domain alone, so the first call opens with evidence. Free scorecard.

  See the page →

• Decision ledger: every decision and what it moved

  A decision ledger records each change, why it was made, what it should move, a frozen baseline and the outcome. Write and grade your first entry free.

  See the page →

• Why did a brand vanish from ChatGPT answers?

  A brand stopped showing up in ChatGPT answers. How to confirm it with a fixed prompt set, find the likely cause, fix it in order and track it from there.

  Read the post →

More guides

• Reporting10 min read

  How can an agency cut month-end reporting time without losing the story?

  Find where reporting hours go, time your own process, automate the pulls and keep the story human. A weekly rhythm and narrative template inside.

  Read the guide →
• Agency growth11 min read

  How do agencies find and win D2C brands that stay?

  Where to find D2C brands worth pitching, how to qualify and audit them from a domain, how to run the first call and price 90 days, and how to keep them.

  Read the guide →

Questions agencies ask