--- title: 'How to choose an AI search optimization agency: 12 questions that expose the pretenders' url: 'https://maw11.preview.mountainairweb.com/blog/how-to-choose-an-ai-search-optimization-agency' markdown: 'https://maw11.preview.mountainairweb.com/blog/how-to-choose-an-ai-search-optimization-agency.md' date: '2026-09-30' description: 'Short answer: an AI search optimization agency is worth hiring if it can show you, in your own server logs and analytics, what changed after it did the work. Everything else is theatre. Nobody can submit a business to ChatGPT, nobody can guarantee a citation, and llms.txt is not a ranking lever. Wh…' taxonomy: category: - 'AI & Search' tag: - ai-search - vendor-selection - geo --- [AI & Search](https://maw11.preview.mountainairweb.com/blog/category:AI%20%26%20Search) September 30, 2026 10 min read # How to choose an AI search optimization agency: 12 questions that expose the pretenders We sell one of these services, so we wrote the questions we would least like to be asked. Use them on everyone, including us. Nicholas Murray Author **Short answer:** an AI search optimization agency is worth hiring if it can show you, in your own server logs and analytics, what changed after it did the work. Everything else is theatre. Nobody can submit a business to ChatGPT, nobody can guarantee a citation, and `llms.txt` is not a ranking lever. What a competent vendor can do is make a site crawlable and unambiguous to the engines, improve the pages that are nearly cited, get the brand mentioned where the engines look, and measure crawler visits, AI referrals and citations month over month. The twelve questions below separate vendors who do that from vendors who resell a tracking tool with a markup and a story. We answer them ourselves at the end, because we sell this too. ## What the market actually sells in 2026 The category has split in two. On one side are **tracking platforms**: [Profound](https://www.tryprofound.com/) (enterprise, with prompt volumes, crawler analytics and an "AI Marketer" agent), [Peec AI](https://peec.ai/) (visibility, sentiment and competitor position across ChatGPT, Perplexity and Gemini), [Otterly.AI](https://otterly.ai/) (seven engines, from $29 a month), and the modules bolted onto the SEO suites, such as [Ahrefs Brand Radar](https://ahrefs.com/brand-radar), which tracks mentions and citations across AI Overviews, ChatGPT, Gemini, Perplexity, Copilot and AI Mode. They all do roughly the same thing: run a list of prompts against the engines on a schedule and record who gets named and who gets linked. On the other side are **services**: agencies and studios that use one of those tools, plus your logs and analytics, to decide what to change on the site and off it. Some of them do real work. Some of them are the people who cold-email "your competitors show up in ChatGPT and you don't" and charge a retainer for vague "AI visibility". As [Surmado](https://www.surmado.com/blog/can-you-pay-to-get-listed-in-chatgpt) puts it, "There is no paid placement system. There is no submission form." That one fact disqualifies a surprising share of the pitches you will receive. Keep the size of the channel in view while you shop. Conductor's [2026 AEO/GEO Benchmarks Report](https://conductor.com/academy/aeo-geo-benchmarks-report/), built on 3.3 billion sessions across 13,000 domains, puts AI referrals at 1.08% of website visits, with 87.4% of that coming from ChatGPT, while AI Overviews appeared on 25% of the Google searches analysed. Previsible's [third AI traffic study](https://www.seo-day.de/news/article/chatgpt-commands-92-of-ai-referral-traffic-heres-what-677?lang=en) (6.77 million sessions, 166 properties) found ChatGPT at 92.4% of standalone-LLM referrals and warned that "one vendor's product decision can halve AI traffic overnight." Small, concentrated and volatile. A vendor who describes it any other way is selling you something. ## Measurement **1. What will you measure, and where does the data come from?** The acceptable answer names three sources: AI crawler hits in your server logs (user agents such as `GPTBot`, `OAI-SearchBot`, `ClaudeBot`, `PerplexityBot`), AI referrer sessions in your analytics (chatgpt.com, perplexity.ai, gemini.google.com and so on), and prompt-level citation tracking from a tool like the ones above. If the answer is only the third, you are buying a dashboard. If it is only "visibility score", ask how the score is built; most are the percentage of a hand-picked prompt list in which you are named, which is easy to inflate by picking easy prompts. **2. Who chooses the prompts, and can we see them?** You should. Prompts should come from your sales calls, support tickets and search data, not from the vendor's imagination, and the list should be frozen for the comparison period so the baseline means something. A vendor who will not show you the prompt list has a reason. **3. What is the baseline, and what does "working" look like in month three?** Expect a written baseline before any change: crawler visits by engine and page, AI referral sessions, citations per prompt per engine. Expect honest targets: more pages fetched by `OAI-SearchBot`, more prompts where you are cited, a first referral from an engine that sent none. Do not expect a traffic forecast. Cloudflare Radar's crawl-to-refer ratios for July 2026, [as reported by eCorp IT](https://ecorpit.com/ai-crawler-crawl-to-referral-ratio-2026-geo-decision/), had OpenAI crawling 251 pages for every visit it referred and Anthropic 1,917, against Google's 4.7. Anyone promising AI traffic in volume is ignoring that arithmetic. ## Method **4. What exactly will you change on our site?** The honest list is short: fix crawler access (robots rules that block `OAI-SearchBot` while allowing `GPTBot` are common and backwards, because [OpenAI's documentation](https://developers.openai.com/api/docs/bots) says sites opted out of `OAI-SearchBot` "will not be shown in ChatGPT search answers"); put the answer in the first third of the page; add comparison tables and sourced statistics; keep the entity data (name, address, locations, identifiers) consistent in structured data across every page; refresh dates on pages that are nearly cited. Our own reading of the 2026 studies is in [what gets a business cited by AI](https://maw11.preview.mountainairweb.com/blog/what-gets-a-business-cited-by-ai). If the vendor's method is mostly "AI-optimized content" produced at volume, ask who edits it and what it is for. **5. What is your position on llms.txt and "AI SEO" plugins?** This is the fastest tell in the whole list. Google's [AI features documentation](https://developers.google.com/search/docs/appearance/ai-features) states: "You don't need to create new machine readable files, AI text files, or markup to appear in these features." Search Engine Journal has [tracked Google's mixed signals](https://www.searchenginejournal.com/googles-llms-txt-guidance-depends-on-which-product-you-ask/575431/) for a year, and the Search team's position has not moved. A good answer is "cheap to ship, harmless, not a lever, and we will not bill you for it." A bad answer is a line item. **6. What do you do off the site?** Citations follow mentions. If the engines never see your brand named on review sites, industry publications, directories and partner pages, no amount of on-site work produces a recommendation. Ask what the vendor does about digital PR, listings and reviews, and whether that is included or a separate engagement. "Nothing, we only do technical" is an acceptable answer if it is said out loud. ## Deliverables **7. What arrives each month, and who is it written for?** You want a report a non-specialist can forward to a CMO: crawler visits by engine and page, referral sessions and what they did, citations gained and lost per prompt, the changes made, and the changes recommended with a reason. Ask for a redacted sample from a current client. If the sample is a tool export with a logo on it, the vendor's value is the markup. **8. Will you show us a client's logs, not just their dashboard?** Any vendor who has done the work has seen `OAI-SearchBot` start fetching a page the week after a robots fix. Ask them to describe one such case in detail, with the client unnamed. The story will be specific or it will be vague. ## Honesty **9. What can you not do?** The answers that should come without hesitation: cannot guarantee a citation, cannot predict what a non-deterministic model will say tomorrow, cannot submit a business anywhere, cannot control what Google does with AI Overviews, cannot make a thin site authoritative with markup. A vendor who answers "nothing, really" has told you everything. **10. Do you use this on yourselves, and can we check?** Type the vendor's own category into ChatGPT, Perplexity and Google AI Mode and see whether they are named. This is not a perfect test, but a vendor who sells AI visibility and has none for its own name has a problem explaining why. ## Ownership and pricing **11. Who owns the data, the accounts and the structured data when we leave?** The analytics property, the tracking-tool account, the schema implementation and any content should be yours or transferable on notice. Ask specifically whether the structured data is generated by a plugin you can keep or by a service that stops working when the retainer does. **12. How are you paid, and what is the contract term?** Three models exist: a fixed audit-and-install, a monthly monitoring retainer, and a performance fee. The first two are fine. The third, applied to AI citations, is a problem on both sides: the vendor cannot control the outcome, so either the fee is on vanity prompts or it is priced to cover failure. Twelve-month minimums before any baseline exists are a sign that the vendor's economics depend on you not measuring. A pilot of three months against a written baseline is a reasonable ask of anyone. ## Claims you will hear, and what to say back | Claim | What to ask | Acceptable answer | |---|---|---| | "We can get you ranked in ChatGPT" | How, given there is no submission or paid placement? | "We cannot rank you. We can make you citable and track whether you are cited." | | "Our AI SEO plugin handles it" | Which engine reads what the plugin outputs, and where is that documented? | A plugin that generates consistent structured data from content is useful; one that promises AI rankings is not. | | "You need an llms.txt file" | Which engine fetches it, and how often did it appear in our logs last month? | "Rarely, if ever. We ship it because it is free; we do not charge for it." | | "Your AI visibility score is 12%" | Of which prompts, chosen by whom, on which engines, on what date? | The prompt list, frozen, with the raw engine outputs available. | | "AI traffic converts 5x better" | On whose site, at what volume, and what is ours now? | Your own referral sessions and conversions, with the caveat that the volume is small. | | "Guaranteed results or your money back" | Define results. | Nobody serious offers this. Walk away. | | "We are an official OpenAI partner" | Show us the agreement. | There is no partner programme that places businesses in answers. | ## Our own answers, briefly We sell [MawGeo](https://maw11.preview.mountainairweb.com/services/mawgeo), so here is how we score against our own list. Measurement: the product logs AI crawler visits by engine and page, detects referrals from AI answers, and tracks citations per prompt per engine, with the prompt list agreed with the client and reported against every month. Method: structured data generated from the page content and kept consistent across locations, robots and crawler-access fixes, and content-structure recommendations that the client or we apply; we do not produce content at volume under this product. Off-site work is not included; we say so and recommend a PR partner where it is needed. Deliverables: a monthly summary written to be forwarded. Honesty: we cannot guarantee a citation and our service page says so in the FAQ. We ship `llms.txt` and, on flat-file builds, a Markdown version of every page, because the cost is nil and some agents use them, not because either moves rankings; John Mueller has called Markdown-for-bots "such a stupid idea" on [Bluesky](https://www.searchenginejournal.com/googles-mueller-calls-markdown-for-bots-idea-a-stupid-idea/566598/), and we think he is wrong about agents but right about rankings. Ownership: the implementation lives in your repository and your analytics, and our contract terms, security posture and vendor-onboarding answers are on the [Trust](https://maw11.preview.mountainairweb.com/trust) page. Pricing: fixed audit-and-install, then a monthly monitoring arrangement, with no performance fees; the engagement models are on the service page. Where we are weakest is question 10: our own name comes up for local developer prompts and only unevenly for broader AI-search prompts, which is roughly what a studio of our size should expect, and is exactly why we track it. ## What to do next Take the twelve questions to every vendor on your shortlist and insist on written answers before a baseline month, not after. If you want a second opinion on what a vendor has told you, or you would rather start with a baseline of your own site, [request a free AI-visibility audit](https://maw11.preview.mountainairweb.com/free-audit); it reports crawler access, entity consistency and current citations, and it is useful whether you hire us or not. ## More from the journal [See all posts](https://maw11.preview.mountainairweb.com/blog) AI & SearchOct 7, 2026 ### What AI search optimization actually means (and what it costs to ignore) GEO, AEO and AI visibility are the same job with different labels. Here is how ChatGPT, Perplexity, Google… [What AI search optimization actually means (and what it costs to ignore)](https://maw11.preview.mountainairweb.com/blog/what-ai-search-optimization-means) AI & SearchSep 23, 2026 ### How to write an llms.txt that AI crawlers will actually use (with examples) A worked llms.txt example for a multi-location business, what the 2026 server-log studies say about who fetches it,… [How to write an llms.txt that AI crawlers will actually use (with examples)](https://maw11.preview.mountainairweb.com/blog/how-to-write-an-llms-txt) AI & SearchAug 19, 2026 ### What actually gets a business cited by ChatGPT, Perplexity and Google AI The 2026 studies are in: schema alone doesn’t move AI citations, llms.txt is barely fetched, and what works… [What actually gets a business cited by ChatGPT, Perplexity and Google AI](https://maw11.preview.mountainairweb.com/blog/what-gets-a-business-cited-by-ai) --- ## Navigation - Parent: [Blog](https://maw11.preview.mountainairweb.com/blog.md) - Previous: [What AI search optimization actually means (and what it costs to ignore)](https://maw11.preview.mountainairweb.com/blog/what-ai-search-optimization-means.md) - Next: [How to write an llms.txt that AI crawlers will actually use (with examples)](https://maw11.preview.mountainairweb.com/blog/how-to-write-an-llms-txt.md)