guide
Generative Engine Optimization: what the data supports in 2026
The long version of this page - what the research supports, what it does not, and how to measure without fooling yourself.
ReadGenerative engine optimizationYour buyer asks AI.
Your buyer asks AI.
It names someone else.
We measure that gap with a fixed panel of your buyers' real prompts, fix what stops AI reading you, and get your brand into the sources it already cites. No llms.txt theatre, no invented hit rates.
3 runs
per prompt
4 layers
access to placements
90 days
minimum honest unit
0
guarantees we cannot keep
panel prompt · run 2 of 3
“Best agencies to build custom AI automation for a mid-size business?”
1named
2named
3named
you
recall 0 / 4
Absent from every recall prompt in the panel. This is the number the audit starts from.
Where brands drop out
Three places you fall out of an answer.
An engine retrieves sources, resolves who you are, then decides who belongs in the set. Miss one and the other two stop mattering.
Retrieval
The crawler never got your content. Blocked at the WAF, or served an empty shell because your app renders in the browser. Nothing downstream can compensate.
Entity
It retrieved you but is unsure who you are. A name that appears three ways, two brands under one legal entity, prices that disagree between pages.
Set membership
You were retrieved and understood, and still left out, because the sources the engine trusts for your category never mention you. The expensive one.
How it runs
Four steps, and the order is not negotiable.
01
Buying-prompt panel
We write the prompts your buyers actually type into ChatGPT and Perplexity and run them repeatedly. Output: a table of who gets named in your niche, how often, and instead of whom. Usually this is the first time a founder sees which competitor AI recommends by default.
02
Retrieval audit
Can AI read your site at all? We check crawler access for OAI-SearchBot, PerplexityBot, ClaudeBot and friends - robots rules, WAF and CDN blocks, rendering. Cloudflare has blocked AI crawlers by default on new domains since July 2025. A lot of 'AI ignores us' cases are literally a firewall setting.
03
Citation source map
For every panel prompt we log which sources the answer cites - the review sites, directories, comparison posts and communities it pulls from. That list becomes your placement target list, mapped from real answers rather than guessed.
04
Before and after, same panel
After the fixes and placements we re-run the exact same panel and show the delta: prompts where you went from absent to named, and where you moved up. Same instrument, same prompts, no moving goalposts.
What we will not sell you
Three things on every competitor deck that do not work.
This category is barely eighteen months old and largely unpoliced. Here is what we checked against primary sources and decided not to charge you for.
llms.txt as the flagship deliverable
An agency that leads its pitch with llms.txt is selling you a file nobody fetches. Ask them for the request logs. We lead with crawler access and citations because those measurably move answers.
Schema markup as magic
Schema is hygiene, not a growth lever. We ship it as part of the entity work because it is cheap and correct. Nobody gets named by ChatGPT because they added Organization markup.
Guaranteed positions in AI answers
Nobody controls what a model says. We set targets on specific panel prompts and show the before and after honestly. Targets, not guarantees. Ask anyone promising a hit rate how they would enforce it.
Pricing
Start with the audit. Continue only if it is worth continuing.
The report is yours either way. If it shows your whole problem is a blocked crawler and one afternoon of work, we will say so and there will be nothing left to sell.
Audit
from $500one-time
delivered in 2 weeks
Panel, retrieval audit, source map
You want to know where you stand before spending anything on fixes.
- Buying-prompt panel for your niche - who AI names, instead of whom
- Retrieval audit: crawler access, robots, WAF and CDN blocks, rendering
- Citation source map: the pages AI cites for your niche
- Prioritised fix list - what to change, in what order, and why
- 60-minute walkthrough call with Ivan
MonitoringMost picked
from $300per month
panel runs on schedule
The same panel, re-run every month
You fixed things, with us or without, and want to see movement over time.
- Monthly re-run of your prompt panel across ChatGPT and Perplexity
- Delta report: prompts gained, prompts lost, competitor movement
- Citation source changes - new sources AI started pulling from
- Crawler access checks so a config change does not silently drop you
- Alert when a competitor displaces you on a tracked prompt
Retainer
from $1,500per month
audit and monitoring included
We do the fixes and the placements
You want the gap closed, not just measured.
- Everything in Audit and Monitoring
- Crawler access and entity fixes shipped on your site
- Placement work: getting your brand into the sources AI cites
- Content targeted at the prompts where you are absent
- Monthly before and after review against panel targets
Sponsored placements, publication fees and paid media are pass-through at cost and always your decision. We never mark them up and we never place them without asking.
Common questions
Asked before booking.
Including the ones where the honest answer costs us the deal.
How do you measure AI visibility?
We build a panel of buying-intent prompts for your niche - the questions your actual buyers ask ChatGPT and Perplexity - and run them on a schedule. For each prompt we record which brands get named, in what order, and which sources the answer cites. That panel is the measurement instrument. The same panel runs before and after the work, so you see movement in the same terms rather than in a vanity dashboard.
How long until results?
Honestly: expect a four to eight week lag between a fix and visible movement in AI answers. AI systems re-crawl and re-weight sources on their own schedule, and third-party placements take time to publish and get picked up. Crawler-access fixes are the exception - once your site is readable, retrieval-based answers can change faster. Anyone promising movement in days is measuring something else.
What if AI can already read my site?
Then the retrieval audit is a short checklist you pass, and the work shifts to the part that matters more: third-party sources. Being readable is the floor, not the strategy. Most of the engagement is then getting you named in the pages AI already cites for your niche.
Why not just do SEO?
Keep doing SEO, this is not a replacement. But SEO optimises for a ranked list of links, and AI answers are a different distribution channel with different mechanics: crawler access rules, citation-source selection, and brand-entity recognition. Your SEO can be healthy while ChatGPT names three competitors and not you. We measure that gap and close it.
Does llms.txt help?
No, and we would rather lose the deal than pretend otherwise. Ahrefs examined 137,000 domains and found almost none of those files were ever requested. Google has said it does not use them. We will add one as ten minutes of hygiene if you want the box ticked. It is not a deliverable.
We are a single-page app. Is that a problem?
Usually yes, and it is invisible until someone checks. Crawlers do not execute your JavaScript, so a client-rendered app can serve a two-kilobyte empty shell to every engine while looking perfect in your browser. If that is what we find, prerendering becomes the first project, because nothing else matters until content exists for a crawler to read.